International Journal of Computer Science Trends and Technology (IJCST) – Volume 8 Issue 2, Mar-Apr 2020

RESEARCH ARTICLE OPEN ACCESS

Descriptive Analysis of Machine Learning and Its

Application in Healthcare
Sai Sruthi Gadde [1], Venkata Dinesh Reddy Kalli [2]
Software Developer [1], Research Scientist [2]
Cardiac and Vascular Group, Medtronic, Bangalore
India
ABSTRACT
The dynamic world of big data in the healthcare sector characterized by huge numbers, complexity, and speeds is
also not suited to conventional research methods. Methods are especially required that can efficiently estimate
models across comprehensive datasets of medical usage data, clinical data, personal computer data, and many other
sources. While these data sets are quite large, they may also be very sparse (e.g., system data may only be accessible
for a small subset of people), creating difficulties with conventional regression models. Most machine learning
approaches successfully overcome these limitations but still are subject to the standard triggers of partiality that are
typical in observatory studies. The models should be tested by standard design tests for researchers using machine
learning techniques like a lasso or ridge regression.
Keywords:- ML

I. INTRODUCTION mysterious, they are, in fact, closely related to

traditional statistical models that most clinicians
The term "machine learning" refers to a large family
recognize. Initially, machine learning was defined as
of mathematical and statistical methods that have
being a system where the work or decisions are made
historically been focused on prediction [1]. We are
automatically from the data instead of the actions
also involved in forecasting healthcare. Which form
being explicitly programmed [4]. This concept is,
of flu is likely to occur in the next influenza season?
however, comprehensive and can cover almost any
How many influenza vials are available to fulfill the
sort of data-driven approach.
care demand? Nevertheless, predictions are not
necessarily the same as predicting drug outcomes.
The job of a doctor is to isolate the effect of an II. DEFINITION OF MACHINE
operation on the outcome of a patient in order to LEARNING
select the correct drug [2]. The same methodological
problems face policy assessments. Some methods of Perhaps it is easier to consider the life of an
machine learning can predict therapy results, and algorithm in a continuum between absolutely human-
some do not. However, in the literature for machine directed and fully machine-led analysis. It is crucial
learning, the gap between prediction and treatment- that you see how much of the structure or parameter
effect estimates is almost entirely absent. In brief, in of a predictive or diagnostic algorithm is said to be an
order to generate highly accurate classification example of computer education [5]. The trade-off
algorithms, a key focus of any machine learning is to between human characteristics of predicational
segment data into training and validation data sets. algorithms against the processing of data is called the
After the algorithms are built, the full data to make machine learning continuum.
the prediction is applied [3]. This is what an algorithm is trying to do. Since
It is no wonder that medicine is overwhelmed by people place fewer expectations on the algorithm, the
groundbreaking claims from machine learning to learning range of the computer is further increased
large-scale healthcare data. Recent examples show [6]. However, a model does not immediately become
that big data and machine learning can build "machine learning," but rather, all such methods exist
algorithms equivalent to human doctors. While
computer education and big data can at first look

ISSN: 2347-8578 www.ijcstjournal.org Page 189

International Journal of Computer Science Trends and Technology (IJCST) – Volume 8 Issue 2, Mar-Apr 2020

in a continuum, based on the number of human detect diabetic retinopathy with a sensitivity that is
constraints put on the algorithm. equal or greater than that of ophthalmologists. This
An example of a high-level machine-learning method model got the diagnosis from the raw pixels of the
in the form of so-called deep learning models has images without any human interference outside a
recently emerged. Deep learning models are team of ophthalmologists who annotated the correct
astonishingly complex neuron networks that have diagnosis on each image [8]. Since the task is
been developed to construct accurate models from mastered with little to no human experience, these
raw data directly [7]. Recently researchers have profound learning algorithms are fundamental in the
demonstrated an in-depth learning algorithm that can Master Spectrum of Learning.

Figure 1. Biological neurons VS artificial neurons network

While less personal guidance is needed, deep lack of data. For the study of empirical evidence,
learning algorithms to recognize images require there are several successful statistical methods [11].
enormous data amounts to capture the full Nevertheless, the sheer quantity of data along with its
complexity, variety, and nuance of real-world features, including unequal data completeness, raises
models. Such algorithms, therefore, often demand the concerns about the potential for new methods of
extraction of the outstanding image features that are addressing issues of treatment efficacy, patient
connected to the result in hundreds of thousands of benefit, strengths of alternative care system models
instances [9]. Higher placement in the continuum of and limitations, policy behavior, etc.
master learning does not mean superiority since Some approaches to machine learning use prediction
various tasks need different levels of human methods based on regression. Lasso approaches, for
involvement. Although spectrum algorithms are also instance, use a correction factor to reduce the overfit
very versatile and can learn several jobs, they often chance. Since Lasso may reduce those variables'
are not interpretable; there is also the velocity aspect coefficients to zero, it is useful for selecting
– the speed at which users can communicate. EMR variables. Most notably, since the regression of Lasso
data is also almost in real-time available. In addition, requires estimating coefficients in a multivariate
data diversity is increasing. Claims and EMR data are model, the use of machine learning in determining
increasingly associated with broad-based health risk the treatment effects is a short step forward [12].
assessments, socio-demographic information, and Many researchers believe that computers do not
vital signs. Some algorithms are used in the easily select the final model specification. This can
optimization of antiurolithiatic activities [10]. More be known. The final model for the theoretical or
recently, new data on human genetic traits as well as clinical plausibility of researchers will, however,
data from devices such as Fitbit and biometric definitely be tested and subject to the standard battery
sensors are available. Such knowledge is vibrant but of design checks. Furthermore, a set of starting
sparse. variables from which the model is built will expertly
That creates challenges for conventional multivariate handle the risk of an unforeseeable scenario [13].
methods like standard smaller-square regression Machine learning approaches make the initial
analysis since many observations are lost due to the variables much more substantial than standard health

ISSN: 2347-8578 www.ijcstjournal.org Page 190

International Journal of Computer Science Trends and Technology (IJCST) – Volume 8 Issue 2, Mar-Apr 2020

care research practice, but the idea of a theoretical or A specific science discipline focused on philosophy,
clinical model must not be thrown out entirely. mathematics, and computer science that aims at
Unfortunately, machine learning protects from the understanding and creating structures that exhibit
typical problems faced by observational data analysis intelligence properties [15].
is simply nothing magical [14]. In fact, it does not Machine Learning
defend against prejudice by merely running machine A sub-discipline of AI in which computer programs
learning methods on Big Data. Increased-sample size, (algorithms) learn predictive power correlations from
for example, is not going to address the bias issue if data examples. The implementation of mathematical
the data collection lacks essential clinical seriousness models on computers is most obviously machine
indicators such as cancer stage in a breast cancer learning. Machine learning uses a wider variety of
model. mathematical methods that are popular in medicine.
New techniques like Deep Learning are based on
models where the underlying information is less
expected and thus capable of processing more
complex data [16].
Deep Learning
Deep learning methods allow a computer to supply
large quantities of raw data to detect or classifying
the necessary representations. Detailed methods for
learning are focused on multiple data layers with
successive transformations that amplify input aspects
of discrimination, which are essential to remove
irrelevant variations. Profound schooling can be
regulated or unregulated. Deep learning approaches
are responsible for many of the new machine learning
advancement [17].

Figure 2. Machine learning is interdisciplinary

Artificial Intelligence (AI)

Supervised Learning
Computer training programs to learn links between data inputs and outputs through analysis of interest outputs
identified by a (typically human) supervisor. After the understanding of correlations, they may predict future

ISSN: 2347-8578 www.ijcstjournal.org Page 191

International Journal of Computer Science Trends and Technology (IJCST) – Volume 8 Issue 2, Mar-Apr 2020

instances based on current evidence. This is one of the best-known fields in machine learning, with many cases in
and outside of healthcare [18].
Unsupervised Learning
Computer programs which learn associations with data without external association concept. In comparison to
simply building upon existing connections, unsupervised learning may classify previously unknown forecasters [19].
Reinforcement Learning
Computer programs that learn behavior by maximizing a given reward. This strategy is inspired by conduct
psychology and was widely used in games where knowledge is ideal, with several potential choices and no specific
worldwide fault costs [20].

III. AI AND DECISION MAKING IN Machine learning builds on current statistical

HEALTH SYSTEMS techniques and uses approaches that are not based on
prior data distribution assumptions, which can be
In fact, efficient health system management is a set of
used for the formulation of hypotheses and
activities of information processing; for example, the
hypothesis testing by using patterns in the data. There
provision of public health or health care.
are also several more variables to be implemented,
Policymakers adjust organizational and governance
generalizable to a much wider variety of data types,
health system structures, funding, and resource
whereas machine-learning models are more difficult
management to achieve health system efficiency and
to understand and can result in more complicated
program objectives [21].
circumstances [24]. Such techniques have been used
The healthcare sector itself requires two main
to test and detect and forecast future incidents in the
processing tasks: the first is to scan for and diagnose
study context. Such implementations are situated in
the historical, review, and investigative, and the
different settings, usually hospitals rather than urban
second is to prepare, execute and follow up a multi-
environments, with consequences for reproducibility
stage mechanism to achieve a potential outcome [22].
and universality in the vast majority of cases based
Hypothesis development, hypothesis testing, and
on data from single centers. Furthermore, both within
intervention constitute the basic form of these
health care and in all information processing
processes in the fields of health system governance
activities in society, the exponential growth of
and treatment. Machine learning can increase the
machine learning continues [25-29].
development and testing of hypotheses within a
health system by exposing previously concealed
patterns in data and thus has the potential for major
implications both at the individual patient level and at
the system level [23].

ISSN: 2347-8578 www.ijcstjournal.org Page 192

International Journal of Computer Science Trends and Technology (IJCST) – Volume 8 Issue 2, Mar-Apr 2020

IV. POTENTIAL EFFECT OF AI ON maker to concentrate on workflow efficiency in a

CLINICAL CARE AND HEALTH way that enables them to use the picture most
effectively and solve several additional cases. The
WORKFORCE same technologies are also required to turn pathology
and other specialties based on image processing [39-
Machine learning has become a "general 41].
technology," which is all-encompassing, can be This means that machine learning produces human
refined over time, and has the potential to produce and computer hybrid systems. Such instances provide
additional innovations. The use of such innovations an ideal combination for the capacity of human
appears to result in "a large economic revolution, beings to produce expectation, cooperate and
with resulting winners and losers." Economists supervise AI systems in order to manipulate AI's
Acemoglu and Restrepo have studied the historical ability to evaluate vast quantities of data in order to
impact of automation – mechanization replacement – identify predictive power correlations or optimize
and claim that automation has been replaced by against a successive criterion [31].
machines in places where machines have a differing
advantage. automation is a relocation effect [26-28].
However, countervailing forces, which increase labor
VI. CONCLUSION
demand, compensate for the effect of displacement: a
growth effect, which increases production and costs.
This allows savings in effect for existing non- In this article, we addressed the direct influence of
automated tasks and for the development of new non- machine learning on healthcare systems but did not
automated tasks, in part involving direct automation examine the indirect impacts of machine learning on
technology. It is worth reviewing the clinical field healthcare systems, the discovery of drugs, and
best currently described in machine literature, others. The prediction is fundamentally difficult:
diagnostic radiography, and seeing if this general technology changes its environment, and the world
trend might relate to health workers [230-33]. produces new possibilities and new technology
constraints. Basically, general intelligence, since a
V. APPLICATIONS OF MACHINE variant of it already exists in human brains, would be
LEARNING AND DEEP LEARNING feasible. However, it seems impossible in the 5-10
years to systematically extrapolate current techniques
Since in-depth learning algorithms developed new
for re-creating general intelligence. However, a
diagnostic image analysis performing norms, some
Federation of "narrow" and "targeted" machine
commentators predicted the eventual retirement of
learning systems capable of solving central health
radiologists and challenged the need for training of
system issues by improving decision-makers' skills
new radiologists. It is possible for machine learners
and thereby setting up new standards in clinical and
to manage more cases and shift responsibility for
management operations can be and therefore should
diagnostic diagnosis to non-radiologists assisted by
be prepared for immediately possible. This is a
machine learning systems as machine learning
tremendous opportunity for the improvement of the
systems function more independently [34-37]. This
health system, as the costs of growing decision-
reorientation of duties will give the healthcare sector
making capability are unlikely to be substantial
an opportunity to reassess the mixture of expertise
across the health sector. There is no other method
and deployment of radiology teams, with more
that can have such an effect without a corresponding
primary care research and less-automatic research
cost scaling. The fixed cost of implementing machine
and unusual cases being treated by less secondary
learning technologies is considerable: the expense of
and tertiary radiologists [38].
research and development and re-tooling a health
The investigators behind a pneumonia-diagnostic
system is significant, but the potential scalability
machinery learning system have established a
means that the investment rationale is
mechanism whereby the technology system first
straightforward. There is an opportunity to grow in
"reads" the image and points to a target for the human
machine learning by creating clinical data sets of
radiologist, thereby allowing a human decision-

ISSN: 2347-8578 www.ijcstjournal.org Page 193

International Journal of Computer Science Trends and Technology (IJCST) – Volume 8 Issue 2, Mar-Apr 2020

high resolution and the appropriate data sharing [10] Gul, Muhammad Tayyab, Ali Sami Dheyab,
frameworks and collaborative work to create both Ekremah Kheun Shaker, Norhayati
productivity and health. Muhammad, and Aslia Natasha Pauzi. "In
vitro evaluation of anti-urolithiatic
REFERENCES properties of Strobilanthes crispus extracted
using different solvents." Research Journal
of Chemistry and Environment. Vol 24
[1] Gulshan V, Peng L, Coram M, et al.
(2020): 1.
Development and validation of a deep
[11] Marcus G. Deep learning: A critical
learning algorithm for detection of diabetic
appraisal. arXiv:1801.00631. 2018.
retinopathy in retinal fundus
[12] Atun R, Aydın S, Chakraborty S, Sümer S,
photographs.JAMA. 2016;316(22):2402-
Aran M, Gürol I, et al. Universal health
2410.
coverage in Turkey: enhancement of equity.
[2] Brand RJ, Rosenman RH, Sholtz RI, et al.
Lancet. 2013;382:65-99. Medline:23810020
Multivariate prediction of coronary heart
doi:10.1016/S0140-6736(13)61051-X
disease in the Western Collaborative Group
[13] Henglin M, Stein G, Hushcha PV, Snoek J,
Study compared to the findings of the
Wiltschko AB, Cheng S. Machine learning
Framingham study. Circulation.
approaches in cardiovascular imaging. Circ
1976;53(2):348-355.
Cardiovasc Imaging. 2017;10:e005614.
[3] Weber GM, Mandl KD, Kohane IS. Finding
Medline:28956772
the missing link for big biomedical
doi:10.1161/CIRCIMAGING.117.005614
data.JAMA. 2014;311 (24):2479-2480.
[14] Stanford University. Algorithm outperforms
[4] Atun R. Transitioning health systems for
radiologists at diagnosing pneumonia
multimorbidity. Lancet. 2015;386:721-2.
[Internet]. Stanford News. 2017. Available:
Medline:26063473 doi:10.1016/ S0140-
https://news.stanford.edu/2017/11/15/algorit
6736(14)62254-6
hm-outperforms-radiologists-diagnosing-
[5] Kocher R, Sahni NR. Rethinking health care
pneumonia/. Accessed: 20 March 2018.
labor. N Engl J Med. 2011;365:1370-2.
[15] Johnson AE, Pollard TJ, Mark RG. 2017,
Medline:21995383 doi:10.1056/
November. Reproducibility in critical care: a
NEJMp1109649
mortality prediction case study. Machine
[6] Badawi O, Brennan T, Celi LA, Feng M,
Learning for Healthcare Conference 2017.
Ghassemi M, Ippolito A, et al. Making big
JMLR W&C Track Volume 68. Available:
data useful for health care: a summary of the
http://proceedings.mlr.
inaugural mit critical data conference. JMIR
press/v68/johnson17a/johnson17a.pdf.
Med Inform. 2014;2:e22. Medline:25600172
Accessed: 20 March 2018.
doi:10.2196/medinform.3447
[16] Celi LA, Moseley E, Moses C, Ryan P,
[7] Jones SS, Heaton PS, Rudin RS, Schneider
Somai M, Stone D, et al. From
EC. Unraveling the IT productivity
pharmacovigilance to clinical care
paradox—lessons for health care. N Engl J
optimization. Big Data. 2014;2:134-41.
Med. 2012;366:2243-5. Medline:22693996
Medline:26576325
doi:10.1056/NEJMp1204980
doi:10.1089/big.2014.0008
[8] LeCun Y, Bengio Y, Hinton G. Deep
[17] Brynjolfsson E, Mcafee AN. The business of
learning. Nature. 2015;521:436.
artificial intelligence. Harv Bus Rev. 2017.
Medline:26017442 doi:10.1038/nature14539
Available: https://hbr.org/cover-
[9] Beam A, Kohane I. Big data and machine
story/2017/07/the-business-of-artificial-
learning in health care. JAMA.
intelligence. Accessed: September 2018.
2018;319:1317-8. Medline:29532063
[18] Helpman E, Trajtenberg M. Diffusion of
doi:10.1001/jama.2017.18391
general purpose technologies. National

ISSN: 2347-8578 www.ijcstjournal.org Page 194

International Journal of Computer Science Trends and Technology (IJCST) – Volume 8 Issue 2, Mar-Apr 2020

Bureau of Economic Research. 1996. No. [30] Adams S. Is Coursera the beginning of the
w5773. end for traditional higher education?Higher
[19] Trajtenberg M. AI as the next GPT: a Education; 2012.[22]Cicchetti D. Neural
Political-Economy Perspective. National networks and diagnosis in the clinical
Bureau of Economic Research. 2018. No. laboratory: state of the art.Clin Chem
w24245. 1992;38:9–10.
[20] Kalli, Venkata & Gadde, Sai. (2020). [31] Cochran AJ. Prediction of outcome for
Technology Engineering for Medical patients with cutaneous melanoma.
Devices - A Lean Manufacturing Plant PigmentCell Res 1997;10:162–
Viewpoint. 9. 1-6. 7.[24]Exarchos KP, Goletsis Y, Fotiadis DI.
10.17148/IJARCCE.2020.9401. Multiparametric decision support system for
[21] Acemoglu D, Restrepo P. Artificial theprediction of oral cancer reoccurrence.
intelligence, automation and work. National IEEE Trans Inf Technol Biomed
Bureau of Economic Research 2018. No. 2012;16:1127–34.
w24196. [32] Kononenko I. Machine learning for medical
[22] Siddartha M. The algorithm will see you diagnosis: history, state of the art
now. New Yorker. 2017;93:46-53. andperspective. Artif Intell Med
[23] Rajpurkar P, Irvin J, Zhu K, Yang B, Mehta 2001;23:89–109.[26]Park K, Ali A, Kim D,
H, Duan T, et al. CheXNet: radiologist-level An Y, Kim M, Shin H. Robust predictive
pneumonia detection on chest x-rays with model for evaluatingbreast cancer
deep learning. arXiv:1711.05225v3 [cs.CV]. survivability. Engl Appl Artif Intell
[24] Golden JA. Deep learning algorithms for 2013;26:2194–205.
detection of lymph node metastases from [33] Sun Y, Goodison S, Li J, Liu L, Farmerie
breast cancer: helping artificial intelligence W. Improved breast cancer prognosis
be seen. JAMA. 2017;318:2184-6. throughthe combination of clinical and
Medline:29234791 genetic markers. Bioinformatics
doi:10.1001/jama.2017.14580 2007;23:30–7.
[25] Bychkov D, Linder N, Turkki R, Nordling [34] Bottaci L, Drew PJ, Hartley JE, Hadfield
S, Kovanen PE, Verrill C, et al. Deep MB, Farouk R, Lee PWR, et al. Artificial
learning based tissue analysis predicts neuralnetworks applied to outcome
outcome in colorectal cancer. Sci Rep. prediction for colorectal cancer patients in
2018;8:3395. Medline:29467373 separate in-stitutions. Lancet 1997;350:469–
doi:10.1038/s41598-018-21758-3 72.
[26] Ein-Dor L, Kela I, Getz G, Givol D, [35] Maclin PS, Dempsey J, Brooks J, Rand J.
Domany E. Outcome signature genes in Using neural networks to diagnose cancer.
breastcancer: is there a unique set? JMed Syst 1991;15:11–9.
Bioinformatics 2005;21:171–8. [36] Simes RJ. Treatment selection for cancer
[27] Ein-Dor L, Zuk O, Domany E. Thousands of patients: application of statistical
samples are needed to generate a robustgene decisiontheory to the treatment of advanced
list for predicting outcome in cancer. Proc ovarian cancer. J Chronic Dis 1985;38:171–
Natl Acad Sci 2006;103:5923–8. 86.
[28] Ayer T, Alagoz O, Chhatwal J, Shavlik JW, [37] Akay MF. Support vector machines
Kahn CE, Burnside ES. Breast cancer risk combined with feature selection for
es-timation with artificial neural networks breastcancer diagnosis. Expert Syst Appl
revisited. Cancer 2010;116:3310–21. 2009;36:3240–7.
[29] Platt JC, Cristianini N, Shawe-Taylor J. [38] Chang S-W, Abdul-Kareem S, Merican AF,
Large margin DAGs for multiclass Zain RB. Oral cancer prognosis based
classifica-tion; 1999 547–53. onclinicopathologic and genomic markers
using a hybrid of feature selection

ISSN: 2347-8578 www.ijcstjournal.org Page 195

International Journal of Computer Science Trends and Technology (IJCST) – Volume 8 Issue 2, Mar-Apr 2020

andmachine learning methods. BMC

Bioinforma 2013;14:170.
[39] Chuang L-Y, Wu K-C, Chang H-W, Yang
C-H. Support vector machine-
basedprediction for oral cancer using four
snps in DNA repair genes; 2011 16–8.
[40] Eshlaghy AT, Poorebrahimi A, Ebrahimi M,
Razavi AR, Ahmad LG. Using threemachine
learning techniques for predicting breast
cancer recurrence. J Health MedInform
2013;4:124.
[41] Exarchos KP, Goletsis Y, Fotiadis DI. A
multiscale and multiparametric approach
formodeling the progression of oral cancer.
BMC Med Inform Decis Mak 2012;12:136.

ISSN: 2347-8578 www.ijcstjournal.org Page 196