See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/371110665

Smart Health Monitoring Using Deep Learning and Artificial Intelligence

Article · May 2023

DOI: 10.18280/ria.370222

CITATIONS READS

5 2,677

1 author:

Dr Suma K G
VIT - AP University
37 PUBLICATIONS 45 CITATIONS

SEE PROFILE

All content following this page was uploaded by Dr Suma K G on 28 May 2023.

The user has requested enhancement of the downloaded file.

 Revue d'Intelligence Artificielle
Vol. 37, No. 2, April, 2023, pp. 451-464
Journal homepage: http://iieta.org/journals/ria

Smart Health Monitoring Using Deep Learning and Artificial Intelligence

Jeethu Philip1 , Suma Kamalesh Gandhimathi2 , Silpa Chalichalamala3 , Balaji Karnam4 , Suresh Babu
Chandanapalli5 , Srinivasulu Chennupalli6*
1
Information Technology, MLR Institute of technology, Hyderabad 500043, India
2
Department of Intelligent Systems, Faculty of Computer Science and Engineering, Lovely Professional University, Punjab
144001, India
3
Department of Data Science, School of Computing, Mohan Babu University, Tirupati 517102, India
4
Department of Computer Science and Engineering, B V Raju Institute of Technology, Narsapur 502313, Telangana State,
India
5
Department of Information Technology, Sheshadri Rao Gudlavalleru Engineering College, Gudlavalleru, AP 521356, India
6
Department of Computer Science and Engineering, Institute of Aeronautical Engineering, Hyderabad 500043, India

Corresponding Author Email: [email protected]

https://doi.org/10.18280/ria.370222 ABSTRACT

Received: 14 February 2023 The genesis and spread of illnesses are a major concern in today's rapidly developing
Accepted: 31 March 2023 technological and evolutionary environment. The prevention and management of illnesses
using technological means have emerged as one of the most pressing challenges facing the
Keywords: medical community. With today's hectic schedules, it's nearly impossible to stick to a
remote healthcare monitoring, Internet of healthy routine. The problems above can be fixed by using a smart health monitoring
Things, deep learning, clinical decision system. Two of the most rapidly emerging technologies are the Internet of Things (IoT) and
support systems, smart city infrastructure artificial intelligence (AI). As more people relocate to urban areas, the idea of a "smart city"
has become increasingly commonplace. Increased efficiency, decreased expenses, and a
renewed emphasis on improving the quality of care provided to patients are central to the
idea of a "smart city." There has to be a thorough familiarity with the various smart city
frameworks before the Internet of Things (IoT) and artificial intelligence (AI) can be
effectively used for remote healthcare monitoring (RHM) systems. Technologies, gadgets,
systems, models, designs, use cases, and applications are all examples of frameworks. The
RHM system, based on the Internet of Things, relies heavily on artificial intelligence (AI)
and deep learning (DL) to analyse the data it collects. However, DL techniques are widely
utilised for making analytical representations, and they are included in CDSS and other
kinds of healthcare services. Patients are given personalised recommendations for therapy,
lifestyle changes, and care plans by clinical decision support systems after each element is
thoroughly analysed. Supporting healthcare applications, this technology may assess
activities, etc. In light of this, this paper presents a survey that zeroes in on the best smart
city applications for the Internet of Things in health. By analysing the most important
monitoring applications across many models using appropriate IoT-based sensors, this
research provides a comprehensive assessment of the technologies and systems involved in
providing RHM services. Finally, this study makes a contribution to scientific
understanding by identifying the primary constraints on this field of study and suggesting
directions for further investigation.

1. INTRODUCTION developments in sensing, processing, spectrum use, and

artificial intelligence. The development of microelectronics
Physical and mental well-being are vital to every human has made feasible these solutions by creating tiny, low-cost
being. Lack of disease and improvement in one's physical and medical sensing devices, which have fundamentally altered
mental well-being are two of the hallmarks of good health, the delivery of healthcare [3]. Therefore, healthcare
which is a great benefit. The health and healthcare systems in organisations divide these approaches into two categories:
every culture are becoming more modern and technologically symptomatic therapy and preventative treatment. People today
advanced. There has been a worldwide economic impact from place a premium on illness prevention, early diagnosis, and the
COVID-19 recently, leading to the implementation of cutting- most effective treatment for a wide range of chronic conditions
edge healthcare technology [1]. Intelligent healthcare systems [4]. As a result, building comprehensive healthcare
use remote nursing of patients to prevent the spread of disease surveillance networks across the country is becoming the norm.
and maximise the efficiency of healthcare spending. An Healthcare has long benefited from technological
excellent option in this setting is the combination of machine advancements, and now those advancements are being made
learning and IoT-enabled healthcare systems [2]. Efficiency in digitally information and smart devices, medical records, and
IoT) and machine learning-based solutions is a result of portable devices create a vast infrastructure in healthcare

451
systems [5]. The advancement of computational methods in extraction from raw telemedicine data. Massive amounts of
healthcare, in tandem with the expansion of physical facilities, diverse data are produced in sports informatics, which calls for
has given healthcare professionals and academics the ability to standardised methods of analysis and information extraction.
create creative solutions across a wide range of previously Without the need for manual engineering or intervention, a
unexplored areas. The purpose of this study was to identify the deep neural network may produce useful features from an
critical elements influencing people's propensity to embrace existing unstructured dataset using typical learning processes.
wearable technologies in healthcare [6]. In order to better This study proposes a real-time, smart healthcare system
facilitate the Internet of Health Things in the industrial setting, that makes use of deep learning algorithms and the Internet of
recent studies have provided a comprehensive overview of the Things. The suggested system employs mobile medical
relevant literature. The research offered a literature assessment devices to collect data from the human body, where it is then
and reported on the current state of big data organization, processed by a variety of deep learning algorithms. There have
analytics, and their technical software design in the healthcare been several advances in the development of health
sector [7]. Patients' records are organised in a standard way so monitoring systems in recent years. Here, we take a look back
that we can reliably file away and retrieve the data we need to at some of the more report on the state of the art in the sector
care for them. Systems of traditional healthcare have been with regards to deep learning and transfer learning. Therefore,
replaced by systems of digital healthcare thanks to smart IoT- the purpose of the proposed work is to provide a
based applications like gadgets, and smart mobile phone comprehensive overview of the current techniques,
healthcare systems. In spite of a patient's geographical location, approaches, and methodologies linked with deep learning and
doctors are now able to easily diagnose and track their transfer learning for health monitoring in order to address
condition because to technological improvements [8]. these shortcomings. Based on the facts presented here,
Technologies have integrated detection and classification of academics will be better able to articulate new ideas for
activities through different sensor modalities, allowing for streamlining healthcare thanks to this review.
autonomous and real-time behaviour analysis, monitoring,
activities of daily living, living, rehabilitation, elderly care,
smart home surveillance, and entertainment. [9]. Smartphones, 2. BASIC KNOWLEDGE OF SMART HEALTHCARE
wearables, and ambient surrounds devices are equipped with a IN CITIES
wide variety of smart sensors for detecting and monitoring
activities, including magnetometers, heart rate monitors, 2.1 Smart city and healthcare facilities
accelerometers, pressure, and wearable cameras. [10]. After
being collected by these sensors, several feature sets are Many non-smart cities want to implement conventional
extracted and transformed using machine learning algorithms healthcare infrastructure in an effort to mimic the notion of
so that human actions may be monitored and classified. These smart city healthcare. To ensure the success of smart cities in
feature sets include frequency domain, temporal domain, and providing residents with quality healthcare services,
wavelet transform. Using deep learning for autonomous objectives of the smart city include ensuring residents have
feature representation has been projected to simplify the need access to high-quality housing and healthcare and fostering an
for manually produced features and improve performance environment that is more agreeable to healthy living. There
accuracy [11]. has to be a system in place to produce new, effective healthcare
Artificial intelligence (AI) techniques like data science solutions.
underpin many of the innovations in today's healthcare system.
Early identification, analysis, management, and presentation
are all facilitated by these. IoT, edge devices, drones, robots,
webcams, and intelligent medical equipment can all be helpful
in a pandemic crisis [12]. In modern healthcare systems that
make use of the Internet of Things, machine learning and deep
learning algorithms are routinely deployed. There is a stronger
connection and outliers in the data produced by these IoT
devices, but there is also a lot of unstructured telemedicine
data. Therefore, in order to lessen the likelihood of redundant
data being collected from athletes, it is necessary to employ
machine learning and deep learning methods to extract useful
characteristics from raw telemedicine data [13]. Wasted
resources and disastrous health outcomes [14] are the results
of storing and processing redundant, lossy, and noisy data in
off-site cloud data centres. When it comes to getting rid of
anomalies and unnecessary information, machine learning and
deep learning algorithms are invaluable. These algorithms
ensure that the management information system only has Figure 1. Rudimentary elements of an IoT-enabled keen city
highly processed telemedicine data available for making healthcare scheme
critical health choices. These algorithms guarantee the optimal
choice is made for delay-sensitive and time-critical As can be seen in Figure 1, the fundamental components of
applications by extracting useful characteristics from massive an IoT healthcare system are realised through the cooperation
volumes of raw data [15]. Deep learning, in contrast to of various systems, architectures, and frameworks. There are
machine learning, utilises several concealed nodes and layers six primary parts to what we call "smart services," and they are
to achieve comprehensive coverage and advanced feature as follows: I smart economics; (ii) smart environments; (iii)

452
smart governments; (iv) smart people; (v) smart mobility; and is a smart service, which is an explicit, complex network
(vi) smart living. The many forms services can take have been design. Virtual visits increased from around 1000 in February
outlined in detail. Life in a smart city is superior in almost 2020 (before the COVID-19 epidemic) to between 3000 and
every way. Moreover, it enables people to participate in 3500 on an outbreak day in April 2020. By providing care
societal activities that directly serve their needs. Citizens while the patient engages in social isolation, telemedicine
employ an array of smart gadgets to interact with and make centres may help decrease the spread of illness during an
advantage of these offerings. When individuals use the epidemic. Case study of machine learning's application to
Internet to provide a large quantity of personal data, the result smart cities is shown in Table 1.

Table 1. Machine learning procedures and use in smart city health care

Machine Learning Application and Use Cases for Smart City Health
Function
Algorithm care
Identifies patterns in data that have not been identified
K-means Health care in the comfort of your own home and
Utilizes a Clustering Algorithm to handle Massive
in the bustle of a metropolitan area
Datasets
As a result of the M-IoT network's ability to Using dbscan, you may see how your unstructured data is
Dbscan process data gathered from human sensors (such arranged.
core body temperature), we will be able to gauge Provides compatibility with a wide range of file types
the efficacy of our communication efforts. Clustering HR databases using conceptual schemas
Controls massive amounts of information by sorting it
into useful categories.
Support Vector Forecasting Healthcare Solutions Human health
Processes massive amounts of data of varying types
Machines pattern, such as real-time detection of blood sugar
Powerful enough for use in sophisticated algorithms for
near
processing data
High-speed data processing and transmission
Both of these techniques are commonly used to extract
features from data.
Clinical Correlation Use in creating medical answers by bringing
CCA draws attention to the connection between the two
Analysis together and combining a wide range of devices,
sets of data through comparison and visualisation.
instruments, and cases
A data source for use in a multi-source anomaly
detection system.
Medical hardware, software, and instruments are
Neural Networks all enhanced by machine learning techniques. It offers a useful learning approach for addressing
Application examples can be found in the field of problems with approximate function evaluation.
rehabilitation administration
Uses unique to monitoring tasks, such as those
Anomaly Detection Statistics on a range of normal and pathological data
involving blood pressure or electrocardiogram
Algorithms Using esoteric data sets for comparison
(ECG) readings. Robotic aid for discovering data
Construction deductions and a binary classifier
outliers

2.2 Healthcare services, requests, and remote sensing these issues. The PRMC is also concerned with data storage
and analysis. The IHP receives real-time data from the health
In a smart city, residents may use services that address monitoring system. The hospital system allows doctors to infer
common problems in their neighbourhood. In terms of a patient's health state based on the HMS report and the
functionality, the hospital are the most relevant. Numerous patient's medical history accessed through the PRMC.
studies and applications have shown that remote sensing can The PRMC is a repository for all patient medical
be useful in the healthcare monitoring sector. Patients' information and data, including current health problems as
electronic internals, such as an (ECG), blood pressure, and recorded in electronic medical records. Headers of limitations
pulse, are detected and collected, and a patient record and protocols are also sent to other linked systems. Local
management centre (PRMC) helps handle and store this data. storage is available inside the health monitoring service. The
Using the patient's linked medical sensor, doctors may monitor patient's medical data and history are kept here, making it a
environmental factors like humidity and temperature and send crucial part of any reduced technological architecture for smart
that data to a central database for analysis and perhaps remote healthcare service. The PC-EHR is a flexible repository for a
therapy. patient's medical history and other identifying information
(name, address, phone amount, etc.).
2.3 Technology and construction of RHM
2.4 Sensing, monitoring, and regulatory
The data for the health nursing facility comes from the
patient's smart device and the caretaker's as well. The (HMS) Intelligent healthcare communities will require sensors,
is the brains of the operation, analysing current conditions and monitoring, and control. Feedback values from these sensors
past data to provide an IHP that may be implemented provide automated monitoring and control for healthcare
immediately. During times of crisis, it can also send out alerts, facilities. The data reported by these sensors allow healthcare
warnings, and special cases. The core components of the practitioners to automate a variety of monitoring and control
intelligent service are the health monitoring service used for procedures. These objectives may be effectively achieved
assessment and supervision, the hospital service used to through the use of the Internet of Things, wireless sensor
acquire diagnosis of health issues, and the rapid response to networks, deep learning, and other technologies. Because of

453
their access to real-time data, smart cities can respond rapidly 3.1 Sensor-based health monitoring system
to the healthcare requirements of large populations. Doctors
and other medical professionals can make effective choices The electronic data stream from the sensors is translated
quickly. The Internet of Things, AI, and computers have into an audible alarm that alerts the patient to his or her current
revolutionised the healthcare industry. Various types of health status. Heart rate, temperature, and electrocardiogram
sensors, such as those used in smartwatches, can be either (ECG) sensors are among the most popular [16]. Sensors for
implanted or worn externally on the human body. measuring temperature, heart rate, and body temperature (Max
Data is delivered via gateways and wireless networks to 30205) were often used in health trackers (BME 680). CO2
reach healthcare providers. Electrochemical glucose sensors sensors are just one type of sensor used in health monitoring
are only one type of implanted sensor that can aid in the systems; others include temperature sensors, RFID readers,
monitoring and management of diabetes. Artificial biochemical detectors like glucometers, and sensors that track
intelligence (AI) gadgets are also used by patients to self- things like body posture and breathing.
manage and tailor their diabetic care. Patients with diabetes
can use these devices to track their glucose levels via sensor 3.2 Smartphone-based health monitoring scheme
data sent to their cellphones. Sleep apnea, rheumatoid arthritis,
intracranial hypertension, and cardiac arrhythmias are all Smartphones are among the world's most valuable tools.
monitored and managed with the use of cutting-edge sensors. There are now around 14 different types of sensors on a
The primary barriers of these tools are in their administration smartphone, and many more are expected to be introduced in
and functioning. It's possible that those in charge lack the the near future [17]. A smartphone's speech monitoring
appropriate training or just neglect to keep the equipment function is a powerful tool. Many different kinds of hardware
charged. To alleviate the burden of managing and operating systems have been created to take use of this adaptability.
the sensors, it is crucial to implement zero-effects technologies Wireless sensors, Bluetooth modules, accelerometers,
(ZETs). A wide range of illnesses, including cardiovascular fingerprint scanners, gyros, magnetometers, barometers,
disease and dementia, may be tracked with the aid of remote proximity sensors, GPS trackers, cameras, and NFC-near field
sensing. To effectively treat these types of diseases, sensors are all part of modern smartphones and are frequently
behavioural monitoring must be implemented. One estimate employed in the creation of health monitoring systems [18].
puts the number of persons living with dementia at 47 million Among the many benefits of smartphones is the large amount
in 2017. According to estimates, that figure will rise to 132 of space they provide for storing data. Primary storage on a
million by the year 2050. Improved remote sensing modern smartphone is more than adequate for archiving the
technologies supported by the (AI) might streamline the patient's information. Smartphones running the Android
monitoring and control essential for treating chronic illnesses. operating system make it simple to access and share
information, manage multiple devices, and communicate in
real time.
3. TAXONOMY ON HEALTHCARE NURSING
SYSTEM 3.3 Microcontroller-based health monitoring system

In this analysis, we have categorised all of the systems All throughout the world, microcontrollers are the most
according to how frequently they have utilised certain popular and widely utilised components of health monitoring
hardware components. So, we've broken down every system systems. Microcontrollers (MCUs) are great for quick
into the following three classes.: processing of raw sensor data. The field of field-
One, systems based on sensors; two, systems based on programmable gate arrays (FPGA) has seen widespread
smartphones; three, systems based on microcontrollers for application for the concurrent processing of massive data sets.
monitoring health. To facilitate communication between the sensors, an MCU
In Figure 2 we see the classification scheme used by the was employed [19].
reviewed intelligent health monitoring system.

4. APPLICATIONS OF IOT AND REMOTE

MONITORING IN HEALTHCARE

4.1 Remote intensive care in home healthcare

The IoT refers to a category of computing devices that

includes devices like transceivers, microcontrollers, and
remote sensors. Health care is prioritised because it is
necessary for humans to oversee, control, detect, and react to
data collected by the system, and costs are reduced as much as
possible without compromising quality. In the last five years,
advanced nations like the United States have been the primary
adopters of telehealth remote monitoring. One reason for this
shift is the rising faith in the efficacy of remote healthcare in
enhancing patient wellness. The results showed that patients'
expectations of success had minimal bearing on their decisions
to utilise IoT for eHealth, which should encourage product
Figure 2. Taxonomy of the reviewed smart health monitoring designers, healthcare providers, and marketers to refine their
schemes

454
approaches to these areas in order to increase adoption. Over architecture, but the systems were not yet ready for
100 remote monitors were carefully placed in the homes of implementation. meeting their shared health services goals
individuals with severe cognitive deficits for the investigations. prior to the initiative. After examining the problems, they
Additional data was gathered to better comprehend patient developed a clinically-based, integrated monitoring system
demographics, treatment regimens, social networks, and centred on primary care.
clinical standing. This study set out to determine if the positive
effects of remote healthcare monitoring on patients differed 4.2 IoT healthcare applications
from those seen in a hospital setting by utilising a control
group that was otherwise identical to the active group except Close monitoring of IoT applications is another feature of
that it did not get remote monitoring. The study showed a IoT services. New methods have been developed to create a
significant drop in the number of patients requiring healthcare system framework that can manage and analyse
hospitalisation (a testament to remote monitoring efficacy). data from electronic health records (EHRs) and medical
Hospitalization rates were higher for the control group (45.5%) images. It's evident that app development is focused on the
than for the remote healthcare monitoring group (35.6%) services required, and it's not a big surprise that sick people
across all diagnoses. When comparing the remote monitor and other users rely on apps. It follows that whereas apps are
group to the control group, the study indicated that the number tailored to the needs of end users, services focus on what the
of weekly nursing visits decreased significantly. developer can provide. Emotion detection in functional
Care for the elderly and the terminally sick might be technologies, accident intensity prediction in severe injuries,
provided in the comfort of their own homes with the help of and effectively evaluating alcoholics are only few of the latest
RPM. As a result, this disproves the hypothesis that severe applications that have employed different ML algorithms in
clinical conditions are difficult to handle via remote healthcare similar ways.
monitoring. Different indicators of detections and Research and development efforts have focused extensively
comprehensive monitoring by an applied totally automated on IoT-driven healthcare systems and technology, including
AD monitoring report have been offered in recent studies. how the IoT may aid with paediatric and geriatric care, chronic
Also, the framework labels people as healthy or ill. The illness tracking, individual wellbeing and activity
incorporation of remote healthcare monitoring into existing management, and more. The numerous pieces of technology,
healthcare system infrastructure is essential. A recent analysis wearables, and healthcare instruments now available on the
found that RPM had been implemented across 13 distinct market were also explored in the research. These novel
healthcare programmes and services in the UK. One chronic healthcare products and tools might be viewed as an example
obstructive pulmonary disease (COPD) patient on oxygen, one of the inventiveness and originality of the Internet of Things
patient with chronic heart failure (CHF), one patient with in delivering various forms of medical assistance. The many
diabetes, one patient in unstable condition (such as heart single- and multi-condition IoT-based healthcare applications
problems, high blood pressure, or irregular heartbeat), and four are discussed in the following sections. Table 2 provides a
COPD-related projects were highlighted in the report. A link synopsis of IoT-based sensors in healthcare applications and
to primary care was included into the models and operational other use cases and deployments.

Table 2. IoT healthcare applications and specific use cases

Technology
Applications Descriptions Specific Use Cases
Applied
Blood Pressure Blood Pressure
A device that detects glucose
levels in the blood without
Maintaining a constant eye
requiring a needle or other
on the patient's blood sugar
Glucose Level intrusive procedure. The system
levels. Connect the patient's IoT, IPv6
Sensing consists of a glucose monitor,
body to the doctor's in real
some form of background
time
processing, and either a phone
call or a computer.
IoT-related ECG checking
Electrocardiogram network comprising Electrocardiogram (ECG) Electrocardiogram,
Monitoring a wireless acquisition and monitoring with a signal- IoT frameworks,
transmission detection algorithm anomaly detection.
device
Possible deployment of Internet
Paediatric and elderly care,
of Things-based services, IoThNet
Imminent chronic
systems, and apps improves framework,
Healthcare disease supervision, private
access to healthcare throughout Intelligent Security
health, and
the world and strengthens data Model
fitness management
security.
It includes integrating a
thermometer with an IoT
channel. The Internet of Things' Used to measure temperature
Body Temperature Thermometer
RFID node acts as and
Monitor Sensors, IoT System
themeasurement of body transmit data in real-time
temperature and also
aid in the transmission of data

455
 Used in shortening the
As part of the rehabilitation duration of the
process, patients' mental health rehabilitation process in
Rehabilitation IoT Frameworks,
may be monitored in real time hemiplegic
System Machine Learning
thanks to a proposed IoT-based patients through integrated
smart rehabilitation network. applications
that can serve as self-care
Ambient Assistant Living
An approach to managing Interventions
Medication medications that uses I2Pack to control medication RFID, IoT
Management and iMedBox to ensure that prescription frameworks,
pills are genuine systems

Create a smart wheelchair

Potential wheelchair design
capable of
with a WBAN
Wheelchair monitoring a patient’s health IoT Frameworks,
integration beaming in sensors
Management Processing prompts and Machine Learning
pivoted by the IoT
relays location
app
data
The oximeter is connected to an
IoT channel
such as Bluetooth or the
Wearables that monitor
Internet to ensure
oxygen levels in critically ill
Oxygen Saturation the real-time transmission of Oximeter, IoT
patients and instantly alert
Monitoring data between system
medical staff to any in
the patient (who may be in a
development.
different
location) and the health
practitioner
Temperature and BP
measurement. Train IoT framework,
Capture data, process, and relay
Smart Devices patients and relay treatment Cloud Computing,
them in realtime
information Big Data
based on data gathered
Diseases involving the heart
Cardiovascular Diseases, and blood arteries are known
Essential Machine learning,
Ambient Assisted as cardiovascular diseases.
Healthcare cloud computing,
Living, Fitness, and Facilitating the performance
Services big data
Neurological of bodily actions such as
walking, running, jogging

4.3 Healthcare implementations using smart device requiring complex management. Patients with severe cases of
requests diabetes can have a wide range of complications. Asthma
attacks, severe respiratory tract obstructions, allergic rhinitis,
There has been a recent uptick in the production of electrical and other nasal-related symptoms might influence the lungs,
gadgets that are controlled by sensors sent from smart heart rate, blood pressure, oxygen saturation in the blood, skin,
smartphones. Because of this shift, smart gadgets are and immune system [21].
becoming the primary forces behind the Internet of Things.
Implementing quality service standards like flawlessness,
dependability, and cross-device portability necessitates 5. RELATED STUDIES
integrating IoT with specialised disciplines [20]. Smart
devices have become a helpful instrument in the delivery of The novel wearables-assisted smart health monitoring with
medical care as a result of the development of a variety of optimum deep learning (WSHMSQP-ODL) model is the work
software applications and hardware items that are compatible of Hamza et al. [22]. For starters, the disclosed WSHMSQP-
with smart devices. In-depth research was conducted, and the ODL method makes it possible for wearables to collect data
characteristics of health apps and how they interact with those on sleep-related activities. The next step is data pre-processing,
of smart devices were laid out in methodical detail [21]. The which involves standardising the data in some way. The
article covered topics including patient-focused apps and WSHMSQP-ODL model makes use of network (DBN) model
health apps in general. Apps that can aid in patient education, to predict sleep quality. To improve the DBN model's ability
training, and information were also explored; these and other to predict sleep quality, we employ the extended seagull
similar apps are referred to as auxiliary requests. optimization (ESGO) technique to fine-tune several of the
The following medical issues might be conveniently model's hyper-parameters. A variety of metrics are used to
checked and diagnosed using the tools already available on analyse the WSHMSQP-ODL method's experimental
smart devices. In addition, technological progress aids medical outcomes. Extensive research reveals that the WSHMSQP-
professionals in making correct diagnoses and providing ODL model significantly outperforms competing models.
effective treatment. It might be used to tell the difference To build HI using SHM data fusion, Moradi et al. [23]
between emergencies requiring immediate care (such as trips presented a semi-supervised deep neural network. Instead than
to the emergency department) and chronic conditions simply being utilised as a quality indicator for HI, prognostic

456
criteria are really being employed as goals for the network the cost of doing so. In this way, the efficacy of the suggested
itself. Here, composite panel fatigue stress was tracked using technique in the field of smart healthcare is defined by way of
the acoustic emission technique, and the resulting feature a case study.
extractions were put to use in the development of a smart HI. Battery life can be conserved during transmission by
The generalised model achieves at least an 81.77 percent score reducing the bitstream burden of the raw ECG signal, as
on the prognostic criteria when tested using the leave-one-out suggested by Sarma and Biswas [27]. A power-aware ECG
cross-validation approach, while the holdout method processing architecture is described, which extracts features in
demonstrates a 77.3 percent improvement in the quality of the real time for continuous health monitoring, and which may
HI. This research shows that a model can still be constructed then be encoded into the bitstream. The suggested architecture
to give HIs that coincide with the expected deterioration investigates the benefits of computationally simple procedures
behaviour even when the real HI labels are unknown but the with a particular emphasis on a sensor node with limited
qualifying HI pattern (according to the prognostic criteria) can resources. The design detects R-peak and P-peak using an
be detected. adaptive thresholding approach and handles two forms of
In order to forecast the system's true operational states, cardiac abnormalities: irregular heart rate variation (HRV) and
Keleko et al. [24] took a data-driven strategy, focusing on the first degree AV block. Validation on the MIT-BIH arrhythmia
Deep Neural Networks' (DNN) multi-class classification for database demonstrates that the proposed method has a
unbalanced data. Although DNNs perform well, there are still sensitivity and positive predictivity of up to 99%. With an
some unanswered problems about how trustworthy they will estimated power usage of 2.11 W and an area overhead of
be as "black box" models in more complicated applications, 0.087 mm2, the suggested design is synthesised at TSMC's 90
especially with regards to the decision-making processes and nm technology node. The minimum operational voltage for
the potential ethical, economical, and transparent this design is 1.2 V, and the maximum operating frequency is
consequences on stakeholders. As for the explanation strategy, 10 KHz.
the Deep SHapley Additive exPlanations (DeepSHAP) For WT status monitoring, Zhu et al. [28] offer a new
techniques are used to both provide reliable findings and offer method based on a convolutional neural network (CNN) that
context for the decision-making process within the DNN. cascades to a long- and short-term memory network (LSTM)
Accuracy, F1-Score, Recall, and Precision are all indicators of and combines kernel PCA (KPCA). To begin, SCADA data
the DNN classifier model's robustness and efficiency, which was filtered using the density-based spatial clustering of
are in turn shown by the resulting framework based on the two applications with noise (DBSCAN) approach. By carefully
primary modules. Last but not least, the DeepSHAP method choosing input variables, the KPCA method was then used to
provides an explanation of the developed model's results, monitor and anticipate faults in WT. This led to the
making it easier for humans to understand, interpret, and trust development of the KPCA-CNN-LSTM model. In conclusion,
the model, thereby increasing the support or stimulation of AI the created prediction model was utilised to analyse various
model applications on large-scale problems, including those in parts of the WT, with the wind farm serving as an example.
the industrial sectors. Verifying the efficacy of the suggested technique,
The problems, services, and uses of smart healthcare experimental results show that the proposed model can aid in
systems are presented by Gupta et al. [25]. In addition, using not only monitoring the condition of the WT, but also in
edge computing and IoT paradigms, offer a CNN-based forecasting the abnormal operating state of the WT at an early
prediction model. Local edge provide for fast resource phase.
availability and response time within the context of edge The HAR system developed by Helmi et al. [29] makes use
computing. Health data gathered by IoT devices is analysed of data collected from wearable sensors in combination with
using a CNN model. Also, edge devices are essential because results from DL and SI applications. Residual convolutional
they allow for fast health-prediction results to be delivered to networks and recurrent neural networks are used to provide a
doctors and patients through edge servers. Accuracy and error lightweight feature extraction strategy (RCNN-BiGRU). Here,
rate are two performance metrics that may be used to evaluate we provide a novel approach for selecting features using the
the suggested process. When compared to other methods, the marine predator algorithm (MPA). Three binary variations of
suggested mechanism has an accuracy of 99.23%. the MPA, designated MPA s, MPA s10, and MPA v, are
By merging several cutting-edge approaches including IoT, created towards this end. To guarantee the quality of their
DT, FoT, CoT, and Blockchain, Manocha et al. [26] create a performance, we put the suggested MPA variations through
smart framework for context-aware monitoring of physical extensive testing, including comparisons to multiple
activity without compromising the confidentiality of the optimization algorithms using various evaluation indicators
healthcare sector. In the suggested research, we use deep and statistical tests. Accordingly, it is determined that MPA v
learning's capacity to process data sequentially to examine an achieved the highest performance when compared to other
elderly person's motions in search of signs of abnormality. MPA variations and other comparable approaches.
Furthermore, the suggested framework can protect an To evaluate the state of bridges in third-world nations, Inam
individual's data using the advanced security features of et al. [30] suggested a two-stage, deep learning-based system
blockchain. The proposed system performed a highly accurate for smart infrastructure organization. First, it uses the
analysis of an out-of-the-ordinary occurrence in a person in Pakistani dataset and the publicly available SDNET2018
real time. The results of the calculations reveal that DT can dataset to find instances of cracking in bridges. The dataset
assist establish successful medical services by connecting photos with cracks were analysed using You Only Look Once
patients and medical care specialists via the use of smart version 5 (YOLOv5). The YOLOv5 s, m, and l models are
healthcare solutions. In addition, the effectiveness of the applied to the dataset in a 7:2:1 split for training, validation,
suggested solution is evaluated in terms of its ability to and testing to calculate the primary indicators (precision, recall,
identify anomalous events, the speed with which models can and mAP (0.5)). All of the models' mAP (Mean average
be trained and tested, the latency of the data processing, and precision) results were compared to see which one performed

457
the best. The results reveal that the YOLOv5 m model parameters in a sophisticated classification scheme. It turns out
outperforms the YOLOv5 s and l models on the test set, with that randomised deep neural networks offer a novel approach
mAP values of 97.8%, 99.3%, and 99.1%, respectively. The to investigating the trade-offs between efficiency and accuracy
U-Net model is used for segmentation of the crack to obtain in practical contexts. In addition, it presents a novel sampling
their precise pixels in the second part of the investigation. approach that eschews padding and uses input frames made up
Pixel width, height, and area are measured and displayed on of the coordinates of 468 face landmarks. To prove the
scatter plots and Boxplots to distinguish between cracks, using robustness of the proposed technique, we compare the
the segmentation mask provided to the attribute extractor. In classification accuracy and training time of the proposed
addition, the results of the suggested YOLOv5 models were randomised classifier with those of a non-randomized version
verified by the segmentation phase. In addition to locating and of the same model and with well-known benchmark models.
grading the cracks according to severity, this research split the The high-dimensional statistical feature matrix is mined by
crack pixels to evaluate their width, height, and area per pixel an auto-encoder-based dimension reduction model, as
in varying lighting situations. It's one of the few research described by Li et al. [34]. The next step is to determine the
efforts to focus on inexpensive bridge health assessments and level of deterioration by determining the canonical correlation
damage detection in poor nations, which often have trouble between the baseline's feature space and the monitoring data
keeping up with routine maintenance and repair of such acquired afterward. In order to quantitatively define a
essential infrastructure. Bridge and similar infrastructure may deterioration process for condition monitoring, a new HI is
be assessed for their state and health on a regular basis using developed from this information. Experimental validation
the suggested model, which can be employed by local agencies results show the suggested HI is able to detect incipient defect
on their way to a smarter and automated damage assessment and is more sensitive to the early stage deterioration process
scheme. when compared to other standard state-of-the-art.
In the study, Naseri et al. [31] construct an IoT-based face By standardising the structural health monitoring data from
mask detector utilising a "Single Shot Multi-box Detector the Wuhan Yangtze River tunnel, Tan et al. [35] create a multi-
(SSD)" and a hybrid deep learning approach. The proposed learning model based on a deep learning algorithm and name
model is novel because of the improvements made in face it GC-GRU to predict the future mechanical behaviours of the
detection and face classification using the developed ASMFO tunnel structure. In order to predict segment strain and opening
by optimising parameters like the threshold in SSD, the over the next 45 days, we used GC-ability GRUs to capture the
number of steps per execution in ResNet, and the learning rate temporal dependencies of past performance and the spatial
in MobileNet, allowing it to be more efficient and to perform correlations among different monitoring data. Other
better than previous models. In this case, an adaptive sailfish experiments were conducted to discuss the model's predictive
moth flame optimization technique is used to optimise the ability, and results were compared to both a single-indicator
parameters (ASMFO). Next, the ASMFO-tuned parameters of prediction model and several well-known classical prediction
a hybrid technique called Hybrid ResMobileNet models like GRU, LSTM, XGboost, LR, and RNN. The results
(HResMobileNet-based classification) are applied to the of the study show that up to a horizon of 20 days, GC-GRU is
identified face pictures in an effort to produce reliable mask the superior model. Using GC-GRU, the learning capacity of
detection findings. Traditional meta-heuristic methods and the model rises by 2.2%, and the predicted errors for joint
current classifiers are contrasted with the proposed mask opening and segment strain are at least 0.02 mm and 8.62 %,
identification model with IoT based on three standard datasets. respectively. Consequently, the GC-GRU model may be used
Thus, the proposed framework's efficacy is measured against to foretell the mechanical behaviours of tunnel constructions.
that of other frameworks and preexisting classifiers by means It is possible to rapidly assess the level of damage to a
of experimental analysis. When compared to SVM, CNN, concrete structure via Ai and Cheng [36] suggested a deep
VGG16-LSTM, ResNet 50, MobileNetv2, and ResNet 50- learning method based on two-dimensional convolutional
MobileNetv2, the accuracy provided by the implemented neural networks (CNNs). The method involves first
ASMFO-HResMobileNet is, respectively. segmenting the EMA signatures into multiple sub-range
Specifically, Kumar et al. [32] suggested a Block chain- responses, then selecting the sub-range responses that
orchestrated Deep learning solution for Secure Data correspond to the maximum indices, in this case root mean
Transmission in an Internet of Things-enabled healthcare square deviations (RMSDs), to build the input of CNNs for
system (hereinafter referred to as "BDSDT"). To guarantee training, allowing for rapid prediction of damage severity
data integrity and safe data transfer, a new, scalable block degree. Crossover studies on a cube-shaped concrete building
chain architecture is first presented, one that makes use of the detecting numerous mass loss damages confirm the validity of
Zero Knowledge Proof (ZKP) technique. Then, to resolve the suggested method. By creating a variety of CNN models,
concerns with data storage costs and data security, BDSDT we can additionally assess the impact of input size on the
incorporates an Ethereum smart contract and the off-chain method's overall performance. The experimental findings
storage Inter-Planetary File System (IPFS). Finally, the show that the suggested method is effective and accurate even
verified information is used to develop a deep learning for small damages, which opens the door to real-world
architecture for HS network intrusion detection. The latter monitoring of concrete buildings.
creates a powerful IDS by fusing Deep Sparse AutoEncoder According to Phan et al. [37], a low-level laser treatment
(DSAE) with Bidirectional Long Short-Term Memory (LLLT) system supported by a deep neural network and the
(BiLSTM). Experimental results on two publicly available medical IoT is proposed. This paper makes significant
datasets (CICIDS-2017 and ToN-IoT) show that the proposed contributions in three areas: (1) it proposes a deep learning
BDSDT surpassed state-of-the-arts in both contexts, achieving model for facial dermatological disorder segmentation using a
accuracy close to 99%. Modified-U 2 Net; (2) it provides a comprehensive hardware
In the study, Di Luzio et al. [33] provide a novel deep neural and software design for an automatic phototherapy system;
architecture for emotion identification using randomised and (3) it develops a synthetic data generation process for the

458
proposed models to address the issue of the limited and In their study, Mahajan and Banerjee [41] use AE signals
imbalanced dataset. In conclusion, we present an LLLT from PLB sources as training data to introduce and evaluate
platform that makes use of MIoT to enable remote healthcare three models, including an artificial neural network and 1D
monitoring and administration. With an average Accuracy of and 2D convolutional neural networks (CNNs). The 2DCNN
97.5%, Jaccard index of 74.7%, and Dice coefficient of 80.6% algorithm is stated to have a 94.79 percent success rate for
on the untrained dataset, the trained U 2 -Net model classifying regions. As for location classification, it was found
outperformed other recent models. Results from our that 2DCNN was the most successful model, with accuracies
experiments showed that our proposed LLLT system of 73.12% and 79.37%, respectively, for determining where
successfully identified facial skin diseases and automatically along the AE source's length the head, web, or foot was placed.
applied phototherapy. Integrating AI with MIoT-based After that, AE signals were generated from real-world rail
healthcare platforms is a major step towards the creation of damage by bending a segment of inverted rail under weights
medical assistant tools in the near future. of 100 kN, 150 kN, and 200 kN. For all loads, the 2DCNN
The work of Soni et al. [38], which uses deep learning model accurately predicted the zone of the AE source, and for
techniques to recognise human physical activity in wearable the loads of higher intensity, it accurately anticipated the
and mobile sensor situations, has also garnered a great deal of position of the AE source along the length (150 kN, 200 kN).
interest. In this paper, we propose a DNN that combines the This study effort hopes that the deep learning technique given
strengths of the convolutional neural network (CNN) and the here will aid in the creation of an AE-based real-time
bidirectional long short-term memory (Bi-LSTM). The monitoring system for rail inspection.
model's efficacy was measured using the WISDM and UCI- In order to facilitate healthcare data analytics and
HAR datasets, both of which are available to the general public. networking for personal health monitoring, Memon et al. [42]
The model has an accuracy of 97.96% in the WISDM and present AiDHealth, an intelligent personal health monitoring
97.150% in the UCI-HAR. The simulation results also framework based on artificial intelligence. When it comes to
demonstrate the superiority of the proposed work over healthcare data analytics, AiDHealth's prediction accuracy is
competing approaches. based on a number of machine learning and deep learning
Using multimodal deep learning and the Tucker models. Research has made use of the large Pima Indian
decomposition, Yu et al. [39] introduce a novel high-order Diabetes (PID) dataset. At which experiments show how well
multi-modal learning approach (F- HoFCM). To improve the the proposed MLPD model works. When compared to other
efficiency of its embedding in edge resources, it also proposes classifiers, AdaBoost's classifiers have the best prediction
a private scheme supported by edge clouds. Parameter updates accuracy. The highest accuracy, 0.975%, was achieved with
using a high-order back propagation algorithm and clustering the AdaBoost classifier. The findings show that the suggested
with a high-order fuzzy c-means are just two examples of the approach is superior to the current best practises. Next, we
kinds of computationally intensive tasks that can be handled train the models to provide a 10-fold cross-validation sickness
efficiently in the cloud. Other processes, like multi-modal data risk index for each sample. In light of these results, it is clear
fusion and Tucker decomposition, are carried out by edge that more extensive trials are required to directly compare the
resources. For privacy, feature fusion and Tucker aforementioned machine learning techniques. AdaBoost and
decomposition are employed since they are nonlinear Decision Tree, both with an AUC of 0.994%, were shown to
processes that cannot be performed on the cloud. The created work together to produce the best forecast. Therefore, the
edge-cloud-assisted private healthcare system greatly AdaBoost classifier can more accurately predict the risk of
improves the clustering efficiency, and experimental findings type 2 diabetes than the present algorithms and pave the way
show that the given technique significantly outperforms the for smarter preventative measures and therapeutic approaches.
existing high-order fuzzy c-means (HOFCM) on multi-modal An EMI model of the embedded SA was initially created by
healthcare datasets. Li et al. [43], and then the association between the SA's
Using Ensemble Deep Learning with Framingham Feature conductance resonant frequency (CRF) and conductance
Extraction (FFE), Sivakumar [40] presents a smart healthcare resonant peak (CRP) and the evolution of concrete strength
system for the diagnosis of the COVID-19 pandemic sickness. was explored. Compressive strength was measured at various
The Maps of pollution hotspots are constructed using data and ages, and conductance profiles were tracked in real time. By
attributes acquired with the help of forecasting techniques. examining the data, it was confirmed that as compressive
Based on the maps, immediate mitigation steps are strength of concrete increases, CRF falls while CRP increases,
implemented to lower air pollution levels, such as storing the as predicted by the EMI model. Empirical equation,
extracted data or feature in a Cloud server. Once implemented convolutional neural network (CNN), hybrid (LR-CNN), and
into patient management systems, this strategy would free up linear regression model (LR) were the four models constructed.
cloud storage once pollution thresholds were reached, The created models use the SAs' conductance as inputs and
therefore reducing pollution from patients' sensors. The Gini provide the specimens' compressive strength as outputs. Four
Index factor information gain strategy is used to pick the most models were developed, and then compared against one
significant characteristics while discarding the least important another. At the same time, the LR-CNN model and other
ones in order to further minimise computational overhead and hybrids were compared across a variety of performance
boost system performance. An ensemble classifier based on metrics. As can be seen from the findings, the LR-CNN model
deep learning is developed to forecast the development of accomplishes quantitative prediction of concrete strength
COVID-19. There will be experimental assessments of growth and shows good performance. Compressive strength of
prediction accuracy, error, precision, and memory for a range concrete may be evaluated and predicted using the suggested
of patient counts. Experimental results show an 8% boost in approach, which is straightforward, precise, quantitative, and
prediction accuracy, a 47% drop in error rate, and a 36% dependable.
decrease in prediction time compared to the state-of-the-art According to a proposal by Nancy et al. [44], deep learning
method. has the potential to revolutionise our ability to quickly and

459
effectively analyse massive amounts of data, gain insightful treated as the testing data and the results are mimicked
knowledge, and solve complex problems. Predicting the onset appropriately. When the findings are replicated, they will also
of illnesses with sufficient accuracy and in a timely manner is serve as training data for the subsequent round of tests. To that
essential for providing early intervention and preventative end, the information gleaned from the Medical Device will
treatment to those at risk. With the increased use of EHRs serve as testing data, and then, once processed, as training data
comes a greater need for accurate prediction models, which for future medical summaries. Metrics such as the proportion
may be achieved by employing recurrent neural network of data sent from the Smart Medical Gadget to the server end,
versions of deep learning that are equipped to deal with the precision of storage, and the efficiency of communication
sequential time-series data. The suggested system collects are used to measure the performance of the suggested solution.
information from Internet of Things devices and applies The simulation results show that the proposed approach,
predictive analytics to electronic clinical data stored in the iCloud Assisted Intensive Deep Learning, may significantly
cloud that is related to a patient's medical history. With an F- enhance healthcare parameters.
measure of 98.86%, sensitivity of 98.8%, specificity of In the system suggested by Rajan Jeyaraj and Nadar [46],
98.89%, and accuracy of 98.86%, an accurate signal prediction and estimate technique is based
As part of their suggested method, Kondaka et al. [45] on a Deep Neural Network. With the help of a smart sensor for
develop a novel algorithm they term iCloud Assisted Intensive signal measurement and a National Instrument myRIO for
Deep Learning (iCAIDL), which helps both healthcare savvy data capture, a prototype of the suggested system has
providers and their patients by utilising an intelligent cloud been developed. The Smart-Monitor incorporates an
infrastructure and machine learning tactics. Collecting intelligent sensor into a home monitoring system. We
preexisting medical information from the data repository is the calculated the accuracy of the proposed Smart-Monitor system
first step in the training phase of an iCloud-assisted intensive in predicting four physiological signals for two users. An
deep learning algorithm. After the data training phase average accuracy of 97.2% was achieved in prototype
concludes, the suggested algorithm, named iCAIDL, begins experimental setup. Because of this, it is clear that the
collecting real-time data from the patient, which is treated as suggested automated system can provide accurate monitoring
testing data and then processed using intense deep learning as intended. The experimental outcomes prove that the
principles before being stored in the Cloud repository. This suggested system is capable of providing trustworthy aid and
allows for sophisticated monitoring of health information and precise signal prediction.
is clear to both physicians and patients. The purpose of a Smart
Medical Gadget is to intensively gather data on a patient's vital
signs, blood pressure, and blood flow in order to create a 6. LIMITATIONS IN SMART HEALTH MONITORING
medical record for the testing phase. The data collected by the
smart Medical Gadget must be transmitted to the Server end In Table 3, we depict for thorough comprehension the
for processing; in this context, "processing" refers to machine potential restrictions, repercussions, and enhancement
learning-based processing, in which the received data is procedures for IoT based smart healthcare systems.
Table 3. Limitations, insinuations, and improvement procedures in IoT smart healthcare applications

Limitations Domain Improvement Process Implications

A small number of IoT gadgets have
no rechargeable options while using a
lot of electricity. As a result,
implementations should prioritise
The transmission of
using low-bandwidth connections.
data through machine
To allow data information transfer via
learning software, Power, Batteries’ source limitation and
the IoT, research power-aware and
devices, and Energy capacity
power-generation techniques.
bandwidth uses a lot
Implementation of software-defined
of electricity.
networks (SDNs) and other
virtualization technologies at a variety
of points along the path to more
efficient power generation techniques.
Decision-making using statistical
Absence of clean and
analogies is often linked to Machine
trained data for
Learning. Data from the past is used to lead to network failure and
Artificial Intelligence Data
make forecasts about the future. The service
algorithms.
ML-based strategy should provide
Tendentious dataset
quick analysis of test data.
Data Learning must feed predictive
Noisy data, dirty analysis
data, and incomplete Data and referencing various labeled trained disruption
data examples.
Utilize domain expertise.
Poor QoS, At the level of individual transactions,
reputation, trust management must be
It prevents incorporating all
Confidentiality, Trust implemented.
possible
packet forwarding Service providers and software
ratio, reputation, engineers need their roles defined.

460
 varying social Point out how certain parts
features, and reviews cannavigate IoT-based systems with
minimal
technicality.
models
Machine learning
There is a need for centralization to
tools are exposed to
enable it
generate, collect, and
to run easily and simultaneously.
process a vast
Data An algorithm must be designed to
amount of
designed
information, which
not to expend extra time tracking
includes unstructured
unimportant data.
and structured data.

6.1 Data security in RHM 7. FUTURE SCOPE AND ITS DEVELOPMENT

Big data or health data refers to the information gathered by These studies show that DL-based MHMS can map raw
RHM networks and IoT devices. High-powered computing equipment data to objectives without requiring a large amount
resources and ample data storage space are needed for of human effort or specialised knowledge. This means that
managing this volume of healthcare data. The solution to the deep learning models may be used to monitor the health of any
problem of managing health care data is found in cloud system, rather than just a select few. Additionally, some
computing and cloud storages. Most of this information, current research tendencies and suggestions for future study
however, is protected by patient confidentiality. The biggest areas are included:
problems in RHM are related to data privacy and security. ❖ Performance of DL-based MHMS is highly dependent
Users, people, and organisations should refrain from misusing on the size and quality of datasets, which is why open-
this information for their own ends. Physical, authentication, source large datasets are essential. DL approaches are
network, computer, and storage security all factor into data backed by extremely sophisticated models. However,
safety. Popularity of cryptography, data encryption, genetic DL models can only go so deep until they hit a wall
algorithms, encipherment, and decipherment techniques Most caused by the size of the datasets. Therefore, the big
privacy and security frameworks provide services through dataset ImageNet with over ten million annotated
untrusted third parties. The banking and financial industries pictures can back up the 152-layer CNN model used as
have recently seen a rise in the popularity of blockchain and the standard for image recognition. When it comes to
the Interplanetary File System (IFS) as a means of safe data MHMS, however, the suggested DL models can
transmission. Blocks of data, each of which records a certain include as much as five hidden layers. Additionally,
number of transactions, are linked together in a chain to form models trained on such massive datasets can be used as
a blockchain. Each new block in the chain is a model seed data for subsequent, more targeted tasks
cryptographically certified addition to the public record of all and datasets. Therefore, it makes sense to create and
prior transactions. Blockchain is a very safe method of share massive databases of machines.
transmitting data since each block includes a timestamp and ❖ Use of Domain Expertise: deep learning is not a silver
the value of the hash of the block before it. There are now bullet for all difficulties encountered in monitoring
functioning blockchain applications in the fields of machine health. Applying DL models to MHM requires
transportation, manufacturing, management, and healthcare. domain expertise for optimal performance. The size of
Some successful frameworks for medical health care services the DL models being used may be decreased by
and SHM networks have been released into the public domain. removing discriminative features, and the final
performance can be increased by using a regularisation
6.2 Limitations of RHM and IoT term that is tailored to the specific job at hand.
❖ When it comes to modelling and data visualisation,
However, these RHM and medical IoT devices do have their deep learning methods—and particularly deep neural
drawbacks, such as the fact that the data they collect is networks—have traditionally been seen as "black
sometimes inconsistent and inaccurate due to the use of boxes," meaning that their underlying processing
different types of sensors. Because of the annoyance that the mechanisms are not understood. Understanding these
wearable sensors, especially those aimed at youngsters, might DL models is made easier by visualisation of the learnt
cause, wireless sensors are preferred. Protecting patients' representation and the applied model, which in turn
private information from hackers is an absolute priority. The makes it easier to construct and tune DL models for
issue of fraud in IoT-based healthcare systems is also complicated machine health monitoring challenges. It
discussed. Overall, smart medical systems benefit chronic has been suggested that we may see the activations
patients more than the usual health care services. Remote areas generated by each layer and the features at each layer
do not have access to the electricity necessary to power of a DNN by means of regularised optimization, and an
Internet of Things devices. Real-time monitoring is made more tSNE model has been presented for high dimensional
challenging in low-powered devices and remote areas due to data visualisation.
the need for a quick and dependable network connection. In ❖ Deep learning that has been transferred from one
other words, these restrictions are not tied to the development domain to another is called transfer learning. This line
of new technologies and may be overcome. of inquiry is instructive in the field of machine health
monitoring since certain issues in this area have
adequate training data while others do not. It is possible

461
 to move machine learning and deep learning models systematic review. Informatics in Medicine Unlocked, 23:
from one domain to another. It is true that there have 100550. https://doi.org/10.1016/j.imu.2021.100550
been some prior research addressing transferred feature [4] Chennam, K.K., Aluvalu, R., Shitharth, S. (2021). An
extraction/dimensionality reduction. The goal function authentication model with high security for cloud
of deep neural networks was modified to include a database. In Architectural wireless networks solutions
Maximum Mean Discrepancy (MMD) metric for and security issues, pp. 13-25.
assessing the difference between the source and target http://dx.doi.org/10.1007/978-981-16-0386-0_2
domains. [5] Panganiban, E.B., Chung, W.Y., Tai, W.C., Paglinawan,
❖ Disparity of Class: Machine data follows a highly A.C., Lai, J.S., Cheng, R.W., Chang, M.K., Chang, P.H.
skewed class distribution in practise, with the majority (2020). Real-time intelligent healthcare monitoring and
of observations falling into a small number of diagnosis system through deep learning and segmented
categories. As an example, in fault diagnosis, the analysis. Future Trends in Biomedical and Health
number of fault data is significantly lower than the Informatics and Cybersecurity in Medical Devices,
number of health data. In order to rectify this inequity, Taipei, Taiwan, pp. 15-25.
many enhanced machine learning models (ELMs) and http://dx.doi.org/10.1007/978-3-030-30636-6_3
SVMs have been developed for machine health [6] Ramana, K., Aluvala, R., Kumar, R., Nagaraja, G.,
monitoring. CNN models with class resampling or Krishna, A.V., Nagendra., P. (2022). Leaf disease
cost-sensitive training, as well as the combination of classification in smart agriculture using deep neural
boot strapping techniques with a CNN model, are two network architecture and IoT. Journal of Circuits,
of the more recent and intriguing approaches that Systems and Computers, 31(15).
explore the use of deep learning in unbalanced class https://doi.org/10.1142/S0218126622400047
situations. [7] Aouedi, O., Bach Tobji, M.A. Abraham, A. (2020). An
With the advent of large machinery data, it is expected that ensemble of deep auto-encoders for healthcare
deep learning will have an increasingly significant future monitoring. In Hybrid Intelligent Systems: 18th
influence on machine health monitoring. International Conference on Hybrid Intelligent Systems
(HIS 2018) Held in Porto, Portugal, pp. 96-105.
[8] Archana, H.P, C., Khushi, Nandini, P., Sivaraman, E.,
8. CONCLUSION Honnavalli, P. (2021). Cloud-based network intrusion
detection system using deep learning. ArabWIC 2021:
RHM has shown great success compared to traditional The 7th Annual International Conference on Arab
health monitoring system which is limited to delay, medication Women in Computing in Conjunction with the 2nd
delay and pre-emption. RHM's integration of cloud computing Forum of Women in Research, Sharjah, UAE.
and blockchain technologies has won the trust of healthcare https://doi.org/10.1145/3485557.3485562
professionals and patients in terms of data security and data [9] Wu, X., Liu, C., Wang, L. Bilal, M. (2021). Internet of
privacy. RHM can meet the specific needs of critically ill things-enabled real-time health monitoring system using
patients and general health monitoring services to achieve deep learning. Neural Computing and Applications,
individual health goals and healthy lifestyles. The wider pp.1-12. https://doi.org/10.1007/s00521-021-06440-6
application of SHM in sports and general maintenance is [10] Kailasam, S., Achanta, S.D.M., Rama Koteswara Rao, P.,
limited and can be further explored with constantly evolving Vatambeti, R., Kayam, S. (2022) An IoT-based
technologies. RHM presents many future opportunities for agriculture maintenance using pervasive computing with
medical services that were previously unattainable. The machine learning technique. International Journal of
present work presents a review of smart health monitoring Intelligent Computing and Cybernetics, 15(2): 184-197.
frameworks that have successfully implemented deep learning https://doi.org/10.1108/IJICC-06-2021-0101
and machine learning approaches and algorithms to achieve [11] Gunanidhi, G.S., Krishnaveni, R. (2021). An
high productivity. Similar review of works on SHM ML and experimental analysis of blockchain enabled intelligent
DL do not report combined SHM-related works. Using IoT in healthcare monitoring system using IoT based deep
RHM has led to fast and responsive remote and real-time learning principles. Annals of the Romanian Society for
monitoring of medical services. Cell Biology, pp. 353-362.
[12] Khanna, A., Selvaraj, P., Gupta, D., Sheikh, T. H., Pareek,
P. K., Shankar, V. (2021). Internet of things and deep
REFERENCES learning enabled healthcare disease diagnosis using
biomedical electrocardiogram signals. Expert Systems, p.
[1] Ali, F., El-Sappagh, S., Islam, S.R., Kwak, D., Ali, A., e12864. https://doi.org/10.1111/exsy.12864
Imran, M., Kwak, K.S. (2020). A smart healthcare [13] Wan, S., Qi, L., Xu, X., Tong, C., Gu, Z. (2020). Deep
monitoring system for heart disease prediction based on learning models for real-time human activity recognition
ensemble deep learning and feature fusion. Information with smartphones. Mobile Networks and Applications,
Fusion, 63: 208-222. 25(1): 743-755.
https://doi.org/10.1016/j.inffus.2020.06.008 https://link.springer.com/article/10.1007/s11036-019-
[2] Kasthuri, S., Jebaseeli, A.N. (2020). Review analysis of 01445-x
twitter sentimental data. Bioscience Biotechnology [14] Panwar, M., Gautam, A., Biswas, D. Acharyya, A.
Research Communications (BBRC), Special Issue, 13(6), (2020). PP-Net: A deep learning framework for PPG-
209-214. based blood pressure and heart rate estimation. IEEE
[3] Bolhasani, H., Mohseni, M. and Rahmani, A.M. (2021). Sensors Journal, 20(17): 10000-10011.
Deep learning applications for IoT in health care: A https://doi.org/10.1109/JSEN.2020.2990864

462
[15] Dash, S., Acharya, B.R., Mittal, M., Abraham, A. [27] Sarma, J., Biswas, R. (2023). A power-aware ECG
Kelemen, A. (2020). Deep learning techniques for processing node for real-time feature extraction in
biomedical and health informatics. Cham, Switzerland: WBAN. Microprocessors and Microsystems, 96: 104724.
Springer International Publishing. https://doi.org/10.1016/j.micpro.2022.104724
[16] Tsakalakis, M., Bourbakis, N.G. (2014). Health care [28] Zhu, A., Zhao, Q., Yang, T., Zhou, L., Zeng, B. (2023).
sensor-based systems for point of care monitoring and Condition monitoring of wind turbine based on deep
diagnostic applications: A brief survey. 2014 36th learning networks and kernel principal component
Annual International Conference of the IEEE analysis. Computers and Electrical Engineering, 105:
Engineering in Medicine and Biology Society, Chicago, 108538.
IL, USA, pp. 6266-6269. https://doi.org/10.1016/j.compeleceng.2022.108538
http://dx.doi.org/10.1109/EMBC.2014.6945061 [29] Helmi, A.M., Al-qaness, M.A., Dahou, A., Abd Elaziz,
[17] Grossi, M. (2018). A sensor-centric survey on the M. (2023). Human activity recognition using marine
development of smartphone measurement and sensing predators algorithm with deep learning. Future
systems. Measurement, 135: 572-592. Generation Computer Systems, 142: 340-350.
https://doi.org/10.1016/j.measurement.2018.12.014 https://doi.org/10.1016/j.future.2023.01.006
[18] Kumar, M.A., Sekhar, Y.R. (2015). Android based health [30] Inam, H., Islam, N.U., Akram, M.U., Ullah, F. (2023).
care monitoring system. 2015 International Conference Smart and automated infrastructure management: A deep
on Innovations in Information, Embedded and learning approach for crack detection in bridge images.
Communication Systems (ICIIECS), pp. 1-5. Sustainability, 15(3): 1866.
http://dx.doi.org/10.1109/ICIIECS.2015.7192877 https://doi.org/10.3390/su15031866
[19] Sütő, J., Oniga, S., Orha, I. (2013). Microcontroller based [31] Naseri, R.A.S., Kurnaz, A., Farhan, H.M. (2023).
health monitoring system. 2013 IEEE 19th International Optimized face detector-based intelligent face mask
Symposium for Design and Technology in Electronic detection model in IoT using deep learning approach.
Packaging (SIITME), Galati, Romania, pp. 227-230. Applied Soft Computing, 134: 109933.
http://dx.doi.org/10.1109/SIITME.2013.6743679 https://doi.org/10.1016/j.asoc.2022.109933
[20] Patel, B., Shah, P. (2021). Operating system support, [32] Kumar, P., Kumar, R., Gupta, G.P., Tripathi, R., Jolfaei,
protocol stack with key concerns and testbed facilities for A., Islam, A.N. (2023). A blockchain-orchestrated deep
IoT: a case study perspective. Journal-of-King-Saud- learning approach for secure data transmission in IoT-
University-Computer-and-Information-Sciences, 34(8): enabled healthcare system. Journal of Parallel and
5420-5434. https://doi.org/10.1016/j.jksuci.2021.01.002. Distributed Computing, 172: 69-83.
[21] Chong, J., Cho, C., Tabei, F, Le-Anh, D, Esa, N., https://doi.org/10.1016/j.jpdc.2022.10.002
Mcmanus, D.D., Chon, K.H. (2018). Motion and noise [33] Di Luzio, F., Rosato, A. and Panella, M. (2023). A
artifactresilient atrial fibrillation detection using a randomized deep neural network for emotion recognition
smartphone. IEEE Journal on Emerging and Selected with landmarks detection. Biomedical Signal Processing
Topics in Circuits and Systems, 8(2): 230-239. and Control, 81: 104418.
https://doi.org/10.1109/JETCAS.2018.2818185 https://doi.org/10.1016/j.bspc.2022.104418
[22] Hamza, M.A., Abdalla Hashim, A.H., Alsolai, H., [34] Li, X., Wang, Y., Tang, B., Qin, Y., Zhang, G. (2023).
Gaddah, A., Othman, M., Yaseen, I., Rizwanullah, M. Canonical correlation analysis of dimension reduced
and Zamani, A.S. (2023). Wearables-assisted smart degradation feature space for machinery condition
health monitoring for sleep quality prediction using monitoring. Mechanical Systems and Signal Processing,
optimal deep learning. Sustainability, 15(2): 1084. 182: 109603.
https://doi.org/10.3390/su15021084 https://doi.org/10.1016/j.ymssp.2022.109603
[23] Moradi, M., Broer, A., Chiachí o, J., Benedictus, R., [35] Tan, X., Chen, W., Yang, J., Du, B., Zou, T. (2023).
Loutas, T.H., Zarouchas, D. (2023). Intelligent health Prediction for segment strain and opening of underwater
indicator construction for prognostics of composite shield tunnel through deep learning method.
structures utilizing a semi-supervised deep neural Transportation Geotechnics, p.100928.
network and SHM data. Engineering Applications of https://doi.org/10.1016/j.trgeo.2023.100928
Artificial Intelligence, 117: 105502. [36] Ai, D., Cheng, J. (2023). A deep learning approach for
https://doi.org/10.1016/j.engappai.2022.105502 electromechanical impedance based concrete structural
[24] Keleko, A.T., Kamsu-Foguem, B., Ngouna, R.H. and damage quantification using two-dimensional
Tongne, A. (2023). Health condition monitoring of a convolutional neural network. Mechanical Systems and
complex hydraulic system using Deep Neural Network Signal Processing, 183: 109634.
and DeepSHAP explainable XAI. Advances in https://doi.org/10.1016/j.ymssp.2022.109634
Engineering Software, 175: 103339. [37] Phan, D.T., Ta, Q.B., Ly, C.D., Nguyen, C.H., Park, S.,
https://doi.org/10.1016/j.advengsoft.2022.103339 Choi, J., Oh, J. (2023). Smart low level laser therapy
[25] Gupta, P., Chouhan, A.V., Wajeed, M.A., Tiwari, S., Bist, system for automatic facial dermatological disorder
A.S., Puri, S.C. (2023). Prediction of health monitoring diagnosis. IEEE Journal of Biomedical and Health
with deep learning using edge computing. Measurement: Informatics, 27(3): 1546-1557.
Sensors, 25: 100604. https://doi.org/10.1109/JBHI.2023.3237875
https://doi.org/10.1016/j.measen.2022.100604 [38] Soni, V., Yadav, H., Semwal, V.B., Roy, B., Choubey,
[26] Manocha, A., Afaq, Y. and Bhatia, M. (2023). Digital D.K., Mallick, D.K. (2023). A novel smartphone-based
Twin-assisted Blockchain-inspired irregular event human activity recognition using deep learning in health
analysis for eldercare. Knowledge-Based Systems, 260: care. In Machine Learning, Image Processing, Network
110138. https://doi.org/10.1016/j.knosys.2022.110138 Security and Data Sciences: Select Proceedings of 3rd

463
 International Conference on MIND 2021, Singapore, pp. [43] Li, G., Luo, M., Huang, J., Li, W. (2023). Early-age
493-503. concrete strength monitoring using smart aggregate
[39] Yu, H., Zhang, Q., Yang, L.T., (2023). An edge-cloud- based on electromechanical impedance and machine
aided private high-order fuzzy c-means clustering learning. Mechanical Systems and Signal Processing,
algorithm in smart healthcare. IEEE/ACM Transactions 186: 109865.
on Computational Biology and Bioinformatics, 28(12): https://doi.org/10.1016/j.ymssp.2022.109865
3100-3109. [44] Nancy, A.A., Ravindran, D., Raj Vincent, P.D.,
https://doi.org/10.1109/TFUZZ.2020.2992634 Srinivasan, K., Gutierrez Reina, D. (2022). Iot-cloud-
[40] Sivakumar, N.R. (2023). Ensemble deep learning for IoT based smart healthcare monitoring system for heart
based COVID 19 health care pollution monitor. disease prediction via deep learning. Electronics, 11(15):
Intelligent Automation & Soft Computing, 35(2). 2292. https://doi.org/10.3390/electronics11152292
https://doi.org/10.32604/iasc.2023.028574 [45] Kondaka, L.S., Thenmozhi, M., Vijayakumar, K., Kohli,
[41] Mahajan, H., Banerjee, S. (2023). Acoustic emission R. (2022). An intensive healthcare monitoring paradigm
source localisation for structural health monitoring of rail by using IoT based machine learning strategies.
section based on deep learning approach. Measurement Multimedia Tools and Applications, 81(26): 36891-
Science and Technology, 34(4). 36905.
https://doi.org/10.1088/1361-6501/acb002 [46] Rajan Jeyaraj, P., Nadar, E.R.S. (2022). Smart-monitor:
[42] Memon, M., Abbasi, S., Rahu, G.A., Magsi, H. (2023). patient monitoring system for IoT-based healthcare
AiDHealth: An AI-enabled digital health framework for system using deep learning. IETE Journal of Research,
connected health and personal health monitoring. CC BY 68(1): 1-8.
4.0. http://dx.doi.org/10.21203/rs.3.rs-2402505/v1 http://dx.doi.org/10.1080/03772063.2019.1649215

464

View publication stats