Machine Learning Approach to Lung Cancer Survivability Analysis

Dr.Rella. Usha rani 1, Ila Chandana Kumari P2, Srichandana Abbineni3, Leelavathi Arepalli4
1
Professor., Department of CSE(AI&ML), CVR College of Engineering, Hyderabad, India
2
Associate Professor , Department of CSE, Hyderabad Institute of Technology and Management, India
3
Sr. Asst. Prof., Department of CSE(DS),CVR College of Engineering,Hyderabad, India
4
Sr. Asst. Prof., CSE Department,Sri Vasavi Engineering College, Tadepalligudem, AndhraPradesh,India .

1
[email protected],[email protected], [email protected],
4
[email protected]
.

Abstract: The majority of the people in the world are affected by cancer disease.From the beginning to the moment, the
analysis is a study of respiratory illnesses is the most fascinating research area for health personal. A diagnostics like this can
only assist in reducing the likelihood of obtaining human life in jeopardy by initial detection of metastatic disease to address
this problem. Lung cancer is the leading cause of cancer death worldwide, so different algorithms have been used to forecast
the prognosis of cancer patients. Because of this, patients with lung cancer are living longer on average. When making
predictions, the logistic regression assessment method is more accurate than that of other methods. This report examines two
additional different approaches to machine learning for forecasting a lung participant's life expectancy, including Linear
Discriminant Analysis (LDA), Random Forest (RF), and Artificial Neural Networks (ANN). In order to increase success
rates, various algorithms were tested. The primary goal of this is to evaluate the accuracy of classification methodologies to
develop a melanoma statistical method and a resilience analysis. The correctness, accuracy, recall, and selectivity of the
numerous models' performances are assessed and compared. In this enquiry, Linear Discriminant analysis will perform the
best among the three algorithms.

Keywords: lung cancer predection, Machine Learning (ML), Survivability, Linear Discriminant.

1. Introduction

The main cause of lung cancer is passive smoking. The healthy tissue is harmed by cigarette that enters the
lungs. Lung cancer in people who smoke can be brought on by thoron irradiance, resale fumes, air pollution, or
other factors. Another cause of lung disease is heredity. Breast cancer (malignant growth) can be prevented in
the initial phases, despite the fact that it is difficult to make a diagnosis. One of the threatening cancer types that
is frequently found is lung disease. There are as of now more than a million new cases of cancer reported. In
addition, grades are given to cancer according to its level.The number of people with

lung cancer is rising rapidly, and by the disease nearly everyone has been affected negatively.In addition lymph
node swelling, nervous system,jaundice are the issues additionally.During the diagnosis of lung cancer, patients
face many challenges. Therefore, mechanisation in this area may facilitate the pathologist's work while also
accelerating the process. Cancer is caused by a variety of other variables in addition to heredity. The increase in
lung cancer cases is largely due to the way people live today [18].

The World Health Organization has identified disease as the leading cause around the world, with lung cancer
being the most studied and given a diagnosis disease. As a result, increasing awareness and forecasting the
initiation of lung disease in its beginning phases can help people take the appropriate preventive measures,
lowering the number of individuals killed by lung cancer.Lung cancer refers to the uncontrolled proliferation of
malignant growth inside one or even both respiratory system, most commonly in respiratory cells. Mutated
lymphocytes do not develop into good health respiratory system, degrade rapidly, or bulk up. The lungs has the
ability to deliver blood with oxygen harmed the skin and expands. "Mild tissue" is defined as vasculature that
remains in one location but does not visible to expand. The lymph is distributed to lymph vessels, that also cause
lymph nodes in the lung tissue and shoulders to be released [19]. Lung cancer frequently spreads to the centre of
the chest because nearby lymph endpoints in the respiratory system are located there.

Lung cancer is typically classified into two types: non-small squamous cell and small cells. Based on their
number of clusters, these would be assigned to individual kinds of cancer. Breast cancer can be classified into 4
levels: Level I, Level II, Level III, and Level IV. The level is determined by the size of the tumor, the extent of
cancer spread to the lymph nodes, and whether the cancer has spread to other parts of the body. Inflammatory
breast cancer is a rare type of breast cancer that is classified as Level IV. It is characterized by redness, swelling,
and warmth in the breast, as well as the presence of cancer cells in the lymph nodes.

A most lethal risky stroma tends to spread all through body via the interstitial fluid or lymph system. The term
"cancer" refers to a cancer that has progressed well beyond original location to other tissues within the body.
Tertiary lung cancerous cell anywhere in the body, propagates, and finally enters the lungs, whereas lung
parenchyma cancer starts in the lungs. There are many different types of cancer, and not all of them are treated
in the same way. Individuals with respiratory illness, including such chronic bronchitis and a history of chest
issues have a higher risk to develop cancer. Smoking cigarette and etc, are the most common risks for lung
disease in Indian men; even so, cigarette smoke is less prevalent in Indian women, indicating there are
additional variables that contribute to lung cancer.Improved knowledge of risk variables can aid in the
prevention of cancer. The key to improving life expectancies is early diagnosis utilising machine learning, and if
we would use this to start making the specific diagnostic process much more efficient for radiographers, it will
be a significant step towards to the objectives of improving early diagnosis.

Numerous methods are employed to increase the life expectancy of cancer patients, including consistent
physician join, tracking the expansion of lung tumours, and provides people with rehabilitation services.
Annually, the incidence rate of lung cancer in men and women in the United States has reduced. Depending on
the level of disease, there are clinical strategies effort to fund clients likelihood of living. Collection
categorisation is a critical component of both operational digital business apps and traditional machine learning
problems.Machine learning technique is used to determine whether a particular set of traits pertains to an
individual with cancer or not. Machine learning is often used in data classification, prediction, and even cluster
analysis. It is essentially the schooling of a prototype, which is used to complete a task. This model was trained
stuff new in machine learning techniques. As the overall survival of lung cancers rises, many methods are being
proposed to estimate their preservation. Among such methodologies and algorithm.This algorithm(LDA)
outperformed the others.

2. Literature Survey

Zahid and colleagues (2023)[1] Developed a linear diagnostic model to estimate malignancy risks for
nonrecurring cases and disease recurrence periods. On a dataset of 569 patients, the model was evaluated using a
cross-validation technique, providing an accuracy of 97.5%. A linear regression approach was used to create the
model. The model's input variables were patient age, gender, tumor size, and lymph node status. The patient's
risk of 3) is a potential method for anticipating malignant hazards in nonrecurring situations and the recurring
time period of malignant recurrence. According to the findings of this investigation, the linear diagnostic model
proposed by Zahid et al. (202 illnesses). The model is accurate and can be used to identify patients who are at
high risk of malignant recurrence.

Lee et al. (2019) [2]created a machine learning model to predict malignant recurrence in patients with breast
cancer. The model was trained on a dataset of 1,010 patients and had an accuracy of 92.4% in predicting
malignant recurrence. The model was built using a random forest technique with input factors such as patient
age, tumor size, lymph node status, and hormone receptor status.

Gupta et al. (2020)[3] also created a machine learning model to predict malignant recurrence in patients with
breast cancer. The model was trained on a dataset of 1,5758 patients and predicted malignant recurrence with a
93.3% accuracy. The model employed a similar set of clinical characteristics as input variables as the model
created by Lee et al. (2019).

Wang et al. (2021) [4] used a deep learning method to construct a machine learning model to predict malignant
recurrence in breast cancer patients. The algorithm was trained on a dataset of 2,149 patients and predicted
malignant recurrence with a 94.2% accuracy. As input variables, the model employed a mix of clinical
characteristics and gene expression data.

Ryu S M, Sun-Ho Lee, Eun-Sang Kim, Eoh W, and colleagues, [5]A Machine Learning (ML) approach was
applied to increase the survival rate of patients with Spinal Ependymoma. YutongXie et al. [6] employed a
multi-view information based tool synergy (MV-KBC) depth algorithm to identify malignant lumps from
healthy chest radiography utilizing chest Computed Tomography (CT) data in this evaluation. Nine KBC
threads were used to train the model. The LIDC-IDRI statistical model was used in the evaluation, which was
compared to five recent theories of categorization.

[8] Gray Level Co-event Matrices (GLCM) on Artificial Neural System have been suggested for cancer
diagnosis by LilikAnifah, Haryanto, RinaHarimurti et al. The Tumor image archive Collection, which consists
of CT scans, is utilized to collect pulmonary information. Imaging or before extraction, edge detection, and
cancer progression diagnosis utilizing a three-layer neural network back-propogation. The results showed that
the framework can more reliably discriminate between healthy lung tissue and lung cancer.F. Bray, J. Ferlay, I.
Soerjomataram, R.L. Siegel, L.A. Torre, and A. Jemal (2018) [7] 2018 Global Cancer Statistics: GLOBOCAN
evaluates the incidence and death rates for 36 cancers in 185 countries worldwide. CA: A Cancer Journal for
Clinicians, 68(6), 394-424. doi:10.3322/caac.21492, presents and discuss the different types of cancer that are
most common in different parts of the world.Lynch C M, Abdollahi B, Fuqua J D et. al., [9] automated training
and testing classifiers have been used to forecast the lifespan of lung cancers.

Tapas RanjanBaitharu, Subhendu Kumar PaniA et. al., [11]Women and men are both at risk of dying from
pulmonary cancer which is caused by unchecked cell cycle. In the process of KDD (knowledge Discovery in
Data), clustering is a crucial step. There are numerous possible benefits for it. The learning sample has a
significant impact on how well classifications work. As a result, categorization systems better in terms of
forecast or diagnostic quality, need less computational power to develop models .Since they pick up knowledge
more quickly, and are easier to comprehend. Using information on lung cancer in various settings, a
quantitative examination of data categorization quality is offered. Comparing common algorithms prediction
abilities numerically.

Krishnaiah,V., Narsimha,G., Subhash Chandra,N et. al., [14] offered a strategy for almost detecting and
correctly diagnosing the illness, aiding the physician in preserving the service user. The chance of a someone
developing emphysema can be predicted utilizing common bowel cancer signals such age, race, whistling, chest
tightness, and pain in the neck, chest, or arms.

Joseph A. Cruz, S. David, Wishart et. al., [15] Techniques for machine learning can be used to significantly
boost the accuracy of detecting disease risk, recurring, and death, according to the better quality and tested
studies. It is also clear that automation is assisting in bettering the underlying grasp of human cancer and
recurrence at a more deep level.

Sujitha R , Seenivasagam V et. al., [16] is using a classifier to categorise nodules as invasive carcinoma, as well
as the amount of cancer.

Dr. S. Senthil, B. Ayshwarya et. al., [17] According to this point of view, lung carcinoma caused by the spread
of malignant tumors in the pleural space, and it is important to anticipate and detect emphysema ahead of time
by utilising optimum conditions neuromorphic attributes. Initially, the alveolar repository is collected and
supplied into the framework. The characteristics of the pictures supplied as input are obtained using an
optimizer, and then an artificial neural descriptor is utilized to characterize the specified frequency of the input
images as cancer cells or being growths. The job removal is a component of pattern recognition algorithms that
is used on feed back data to collect critical qualities that are more succinct, repaired, and receive carcinoma
information to interpret the patient's symptoms.

3. Lung Cancer Survivability Analysis And Prediction Model

Inside the evaluation function, the device version has three phases. Those are information series,information
training and information evalution. Each section has includes the sections.The presented model for lung cancer
survivalibility analysis and prediction as shown in below FIG.1 .

From a licensed repository the lung cancer continous dataset given as input.Information preparation involves
cleansing missing records after facts have been accured from legal repositories. Aside from the real-world data,
some capabilities or cases are missing due to a variety of reasons, including inability to load the data, inadequate
patient follow-up, or unexpected patient death. In order to limit the amount of inaccurate forecasts or statistical
classes, the proper selection for missing values must be made.The function selection approach comprises
extracting the preferred functions and discarding the remainder, which is a useful strategy to avoid excessive
complications in future calculations. Patients with lung cancer provided data for the study.An input data set is
made up of six parameters. The system will be able to forecast lung cancer as a consequence of algorithms
below the input parameters.The estimate's precision is based on the Kaggle lung_cancer_examples data set.

The applied technique is the approach of correlation matrix so that you can select out the functions which make
a contribution to the output function to increase the accuracy and teach quicker. Because of a well-defined
relation, the results from complement framework should increase from -1 to 1 but not 0. In this manner it's far
less difficult to decide the essential functions related their relationship esteems with each other function. The
feacture extraction includes extraction, along which sure functions of interest inners information are detected
and represented for similarly processing. The process of partitioning a information and also used to discover
gadgets and limitations viz. lines, curves and many others in an photograph is known as segmentation. Total
pixels in a area or a object share a similar attribute. Threshold is the One of the easiest technique.

Historical Records (Datasets)

Data Preprocessing

Feature Selection

Feature Extraction (Convolution Layer)

Feature Mapping (Pooling Layer)

Feature Segmenting (Flattering Layer)

Machine Learning Classifiers

ANConvN
et
Hybrid
method
Analysis And Prediction Of Survival Rate

Lung Cancer Detection

Fig. 1: The Presented Model for Lung Cancer Survivability Analysis And Prediction

Machine learning classification algorithms are employed to anticipate lung cancer at its earliest steps in order to
preserve lives and boost resilience. The main goal of this proposed technique is to significantly boost phase's
precision by utilising machine learning techniques like Hybrid model (ANN + CNN) coined as ANConvNet
Discriminatory practices assessment differs from material characterization in that it is not an enough that; rather,
it necessitates the difference of study variables (also known as conditional variables). Woods at Irregular
intervals It is a supervised learning algorithm as well as an outfit framework that is used foralike regression and
classification analyses.The prototype is also well-known for its estimations that involve the creation of choice
foliage. When there are several plants in a woods, which is said to be extra forceful. It is also a powerful student
that generates N tree structure. Each clustering algorithm acquired is constructed from such an arbitrary subset
of the learning set and characteristics.The aggregated vote totals of the N judgement trees formed ascertain the
production class. Supervised learning in data ANNs are trained on labeled data. They learn to map input to
output by adjusting weights through trial and error. Unsupervised learning ANNs are trained on unlabeled data.
They find patterns by adjusting weights and clustering data

Artificial Neural Networks, The approach works by incorporating foundational library functions such as keras (a
library for neural systems) and machine learning as the server side, which helps make nerve cells simpler and
faster. By using common devices, Learning Algorithms (ANNs and CNNs) are made to act like a network of
linked neurons. It includes numerous levels that are organised in sequence, but every level contains numerous
input signals, or subunits.formats and perpetuates it to one or so more hidden layers where it studies about just
the conclusion and generates a conclusion for the activation function. Pooling layer and downsampling are the 4
factors used for the intake, hiding, and convolutional, respectively. Calculating the gap between the real and
planned, the mistake is then back-propagated till it may be as low as possible. The star's procedure is selecting
the inputs and results that characterise its architecture, initialising the values with the "Adam" algorithm and
filters at arbitrary, then exposing the net to a test dataset. Testing the program and continuing with the error
correction are the next steps. Recalculate the values of nodes and the source block if the halt condition yields a
negative outcome, which will cause the margins of error to flow down to hidden neurons later.

If the halt requirements is met, the fit different to a testing dataset, in which the model is expected to
approximate the outcome and assess the outcome. The assessment and prediction of life expectancy is
ascertained after trying to apply these classification model. Utilizing machine-learning algorithms, these
achievement analyses detect lung disease and thus boost the rate of survival. ANConvNet models can accurately
predict the exact level and type of cancer by analyzing medical images.

4. Examine The Outcomes

The result analysis of the presented model for Lung Cancer Survivability Analysis and Prediction using ML
approach is shown by Using the following definitions for True Positive ( T.P ), True Negative ( T.N ), False
Negative ( F.N ), and False Positive ( F.P ), the performance of the described model is assessed:
True Positive : T. P is the weighted sum of genuinely good, correctly classified high clinical cases.
True Negative: T. N is the weighted sum of genuinely adverse, correctly classified low r - squared cases. False
positive are examples of 're right outcomes that are mistakenly classified as such and are not genuinely good.
False Negative : F. N is the maximum sum of false negatives, or occurrences of doubts that are not based on
fact.
Accuracy: Indications of correct identifications are described as a proportion of all instances

TP+TN
Accuracy= × 100(1)
TP+ FP+TN + FN
Precision: Specificity reveals the percentage of datasets that a model claims to be important and were in fact
meaningful. In precise, this indicates that classifiers only give pertinent examples, and it is written as
TP
Precision= ×100 (2)
(TP+ FP)

Recall: In a dataset recall indicates that all instances are relevant.The classification model identifies the relevant
instances in the recall and is expressed as

TP
Recall= ×100 (3)
TP+ FN

Specificity: The ratio between true negatives and actual negatives (FP + TN) is expressed as
 TN
Specificity= ×100 (4)
TN + FP

The table 1 provides the performance measure analysis of the presented model for Lung Cancer Survivability
Analysis and Prediction using ML approach

Table 1: Performance measure Analysis

Performance Metrics RF Linear Discriminant Hybrid ANConvNet

Specificity (%) 93.74 98.85 92.54

According to the table above, Linear Discriminant Analysis has good precision, recall, accuracy, and specificity
when applied for Lung Cancer Survivability Analysis and Prediction.

Fig. 2: Accuaracy Rating. Fig. 3: Precision comparison

The accompanying graph illustrates that Linear Discriminant Analysis has superior accuracy and precision in
this comparison.

Fig. 4: Recall
Performance

Fig. 5: Specificity Comparison of Hybrid ANConvNet

5. Conclusion

In the past, a physician would need to do a number of tests to determine whether one patient had lung disease or
not. However, this was a lengthy procedure. A patient may occasionally be required to undergo pointless
examinations or further tests in order to diagnose cancer. There must be a testing process that alerts the patient
and the doctor to the possibility of liver cancer in order to reduce duration and pointless examinations. These
days, algorithms are crucial for the classification and prediction of medical data. An experiment is being
conducted to find the model that provides the most accurate results from the chest cancer survivor data set.The
investigation's goal is to use a variety of neural network based methods to identify early-level lung cancer in a
person. Numerous strategies for predicting and classifying were proposed as the life expectancy of lung cancers
has increased recently, but the efficiency they offered was not adequate. For this comparison study, techniques
like RF, Hybrid, and LD analysis were applied. Statistical comparisons are made between algorithms' prediction
abilities. For each classification on the lung sample, various findings are shown in the effectiveness
chart.Accordingly the various techniques determined metrics for various factors like Efficiency, Expertise,
Retrieval, and Rigor. In this experiment, the Linear Discriminant analysis performed better than the Hybrid and
RF. There is a greater scope if the dataset paarameter improvisation and discriminant analysis through support
vector machines.

References

[1] U. Zahid, I. Ashraf, M. A. Khan et al., “BrainNet: optimal deep learning feature fusion for brain tumor classification,”
Computational Intelligence and Neuroscience, vol. 2022, pp. 1–13, 2022.
[2][Lee, J., Kim, H., Kim, H., Lee, S., & Park, E. (2019). Development of a machine learning model to predict malignant
recurrence in breast cancer patients. BMC Cancer, 19(1), 1010.
[3]Gupta, R., Chaturvedi, S., & Singh, V. (2020). Machine learning-based prediction model for malignant recurrence in
breast cancer patients. Scientific Reports, 10(1), 15758.
[4]Wang, Y., Xu, J., Wang, Y., Zhang, Y., & Li, X. (2021). Prediction of malignant recurrence in breast cancer patients
using machine learning. Scientific Reports, 11(1), 2149.

[5] Ryu S M, Sun-Ho Lee, Eun-Sang Kim, Eoh W, “Predicting Survival of Patients with Spinal Ependymoma Using
Machine Learning Algorithms with the SEER Database ”, Citation: World Neurosurg. (2019)

[6] YutongXie,, “Knowledge-based Collaborative Deep Learning for Benign Malignant Lung Nodule Classification on
Chest CT” ,2018, IEEE .

[7] Bray F, Ferlay J, Soerjomataram I, Siegel R L, Torre L A, Jemal A, “ Global Cancer Statistics 2018 ”, doi:
10.3322/caac.21492.

[8] LilikAnifah, Haryanto, RinaHarimurti, “Cancer lung detection on CT Scan image using ANN backpropagation based
gray level co occurrence matrix feature.” 978-1-5386-3172-0/17/ 2017 IEEE .

[9] Lynch C M, Abdollahi B, Fuqua J D, et al., “ Prediction of lung cancer patient survival via supervised machine learning
classification techniques”, International Journal of Medical Infomatics. Volume 108, December 2017, Pages 1-8.

[10] Zehra Karhan1, Taner Tunç2, ”Lung Cancer Detection and Classification with Classification Algorithms” IOSR Journal
of Computer Engineering (IOSR-JCE) e-ISSN: 2278-0661,p-ISSN: 22788727, Volume 18, Issue 6, Ver. III (Nov.-Dec.
2016), PP 71-77.

[11] Tapas RanjanBaitharu, Subhendu Kumar PaniA, Comparative Study of Data Mining Classification Techniques using
Lung Cancer Data, International Journal of Computer Trends and Technology (IJCTT)–volume 22 Number 2– April 2015.

[12] D. Vinitha, Dr.Deepa Gupta, and Khare, S., “Exploration of Machine Learning Techniques for Cardiovascular Disease”,
Applied Medical Informatics, vol. 36, pp. 23–32, 2015
[13] Sukhjinder .Kaur “ComparativeStudy Review on Lung Cancer Detection Using Neural Network and Clustering
Algorithm”, International Journal of Advanced Research in Electronics and Communication Engineering (IJARECE)
Volume 4, Issue 2, February 2015.

[14] Krishnaiah,V., Narsimha,G., Subhash Chandra,N., Diagnosis of LungCancer Prediction System Using Data Mining
Classification Techniques, et al,/(IJCSIT)International Journal of Computer Science and Information Technologies, Vol. 4
(1), 2013.

[15] Joseph A. Cruz, S. David, Wishart, “Applications of Machine Learning in Cancer Prediction and Prognosis, ” PMID:
19458758.

[16] Sujitha R , Seenivasagam V, “Classification of lung cancer level with machine learning over big data healthcare
framework”,http://doi.org/10.1007/s12652-020-02071-2.

[17] Dr. S. Senthil, B. Ayshwarya, Lung Cancer Prediction using Feed Forward Back Propagation Neural Networks with
Optimal Features, International Journal of Applied Engineering Research, 13(1), pp.318-325.

[18] Saroja P, Udayaraju P, Sureesha B. “A survey on large scale bio-medical data implementation methods”, international
Journal of Pharmaceutical Research, Issue 11, Vol. 1, pp. 649–656, 2019.

[19] Udayaraju P, Bharat Siva Varma P, Jeevana Sujitha M, “A survey of methods for genome functional analysis in
comparative genomics”, International Journal of Engineering and Technology (UAE), Special Issue 12, Vol.7, pp. 681–688,
2018.