IJARSCT

International Journal of Advanced Research in Science, Communication and Technology

International Open-Access, Double-Blind, Peer-Reviewed, Refereed, Multidisciplinary Online Journal

ISSN: 2581-9429 Volume 5, Issue 1, October 2025 Impact Factor: 7.67

Lung Cancer Prediction Using Machine Learning:

A Comparative Analysis of KNN, SVM, Random
Forest, and Logistic Regression
Emmanuel Ifeoluwa Oyerinde, Abosede Ibironke Ojo, Adedoyin Samuel Adebanjo,
Aisha Omorinbola Ajao, Afoluwajuwonlo Obaoye, Opeyemi Adelowo, Royce Nwoko,
Emmanuel Alexander, Chidalu Chukwudebelu
Department of Information Technology
Babcock University, Nigeria
oyerindee@[Link]

Abstract: Among the major issues of cancer-associated fatalities universally is lung cancer, and survival
rates are heavily reliant on prompt and precise diagnosis. Conventional diagnostic techniques, while
effective, frequently miss early-stage lung cancer detection. By examining intricate patterns in medical
data, machine learning offers a reassuring method for enhancing the prediction of lung cancer. However,
the algorithm and optimization strategies employed determine how successful machine learning models
are. This study explores the comparative evaluation of four machine learning approaches for the
prediction of lung cancer: Random Forest, Logistic Regression, K-Nearest Neighbors (KNN), and
Support Vector Machine (SVM). To improve model performance, the Kaggle dataset was preprocessed,
encoded, and put through feature selection procedures. Hyperparameter tuning was used to refine model
parameters acceptable to upsurge accuracy still more. Key performance pointers counting accuracy,
precision, recall, and F1-score were accustomed to evaluate the models. The findings demonstrate with
an accuracy of 90%, the Logistic Regression method performed best, with other models exhibiting varied
degrees of performance. The outcome of this work highlights the significance of model assortment and
parameter optimization, as well as the promise of machine learning in the prediction of lung cancer.
Future research could explore deep learning approaches and integrate additional patient data to
enhance predictive performance. Ultimately, leveraging machine learning for lung cancer diagnosis
could lead to earlier detection, better persevering consequences, and a significant decrease in death
rates.

Keywords: Lung cancer, Machine learning, Random Forest, Logistic Regression, Hyperparameter tuning

I. INTRODUCTION
The lungs are essential respiratory organs in charge of gas exchange, particularly the elimination of carbon dioxide and
the absorption of oxygen, which sustains cellular metabolism in the human body [1]. Lung cancer, a malignant
proliferation of lung tissue, remains one of the most lethal diseases worldwide, with increasing mortality linked to
smoking, air pollution, and exposure to carcinogens. The disease often progresses silently in early stages, manifesting
symptoms such as chronic cough, hemoptysis, chest pain, and fatigue only when it has advanced [2].Machine Learning
(ML) is asubset of artificial intelligence permitselectronic devicesto absorb from data without the need for explicit
programming [3]. It has found transformative applications across domains, from personalized recommendation systems
to medical diagnostics. In healthcare, ML has proven especially promising in analyzing high-dimensional datasets and
extracting non-trivial patterns for predictive modelling [4].
For illness classification tasks, including cancer prediction, machine learning techniques together with K-Nearest
Neighbors (KNN), Support Vector Machines (SVM), Random Forest Classifier (RFC), and Logistic Regression (LR)

Copyright to IJARSCT DOI: 10.48175/IJARSCT-29107 53

[Link]
 IJARSCT
International Journal of Advanced Research in Science, Communication and Technology
International Open-Access, Double-Blind, Peer-Reviewed, Refereed, Multidisciplinary Online Journal

ISSN: 2581-9429 Volume 5, Issue 1, October 2025 Impact Factor: 7.67

have been used. Every method presents different compromises between interpretability, computational effectiveness,
and performance in categorization.
1. KNN is a straightforward and flexible method that classifies new data points according to the common vote of
the adjacent neighbors in feature space [4] [5].
2. SVM looks for the best hyperplane to separate classes, demonstrating high accuracy in jobs involving binary
categorization, including identifying instances that are malignant and those that are not [6].
3. A variety of decision trees are built using Random Forest, an ensemble learning approach, to increase
resilience and reduce overfitting [7].
4. A sigmoid function is employed in logistic regression to represent the probability of a binary outcome and is
favored for its interpretability in clinical settings [8].
Despite the success of these models, challenges persist, such as obtaining high-quality annotated datasets, tuning
hyperparameters, addressing model overfitting, and deploying models in clinical workflows [8]. This study conducts a
comparative analysis of KNN, SVM, RFC, and LR using publicly available lung cancer datasets to assess their
predictive performance through system of measurement like exactness, correctness, precision, recall, and F1-score.
With the intention to improve data-driven decision and improve patient results, the ultimate objective is to assess the
viability and clinical usefulness of machine learning techniques in early lung cancer detection.

A. The Lungs’ Anatomy and Function

The thoracic cavity contains the lungs, which are spongy, cone-shaped organs. Because of the heart's restriction of
space, the left lung has two lobes while the right has three [1]. They perform gas exchange approximately 12–20 times
per minute, a process crucial to sustaining life. Protective mechanisms such as nasal hairs, mucus lining the airways,
and the sweeping motion of cilia work collectively to filter airborne pollutants [1].

B. Lung Cancer: Overview and Classification

Lung cancer arises from epithelial cells and is separatedinto two keyclasses: small cell lung cancer (SCLC) and non-
small cell lung cancer (NSCLC). NSCLC, which comprises big cell carcinoma, squamous cell carcinoma, and
adenocarcinoma, makes up around 85% of cases. More aggressive, SCLC grows and spreads quickly, and is usually
presented as restricted or vast [2][9].
Major risk factors include tobacco use (accounting for ~90% of cases), secondhand smoke, occupational exposure (e.g.,
asbestos, arsenic), air pollution, genetic predisposition, and emerging risks like vaping [2][10]. Notably, early-stage
lung cancer often lacks symptoms, making early detection critical to improving survival, which remains around 19%
overall [2].

C. Predictive Analytics in Medical Diagnosis

Using both historical and current data, predictive analytics makes predictions about the future outcomes using statistical
models, ML, and AI [11]. In oncology, predictive modelling can identify patients at high risk, guide diagnostic testing,
and inform personalized treatment strategies.

D. Machine Learning Paradigms

ML algorithms are categorized into four main paradigms:
1. Supervised Learning: Uses labelled data to train classifiers (e.g., SVM, decision trees, KNN, LR) for tasks
like cancer diagnosis [12].
2. Unsupervised Learning: Uses methods like dimensionality reduction and clustering to find patterns in
unlabeled data [13].
3. Semi-Supervised Learning: To increase learning efficiency, a small labeled dataset is combined with a larger
unlabelled sample [14].
4. Reinforcement Learning: Involves learning optimal actions through feedback mechanisms, commonly used
in robotics and adaptive systems [12].
Copyright to IJARSCT DOI: 10.48175/IJARSCT-29107 54
[Link]
 IJARSCT
International Journal of Advanced Research in Science, Communication and Technology
International Open-Access, Double-Blind, Peer-Reviewed, Refereed, Multidisciplinary Online Journal

ISSN: 2581-9429 Volume 5, Issue 1, October 2025 Impact Factor: 7.67

E. Review of Related Works

[14] sought to weigh the efficiency of diverse old-fashioned machine learning methods for example SVM, Decision
Trees, and Random Forests in identifying lung cancer through clinical data analysis. The researchers made use of a
dataset containing patient demographics, clinical characteristics, and histopathological information. They utilized a
methodical strategy for selecting features in order to improve the algorithms' effectiveness. The research included
thorough testing and verification to confirm the accuracy of the findings. SVM outperformed other algorithms with an
accuracy of 92% and a sensitivity of 90%, suggesting its superior effectiveness. The Study highlighted the significant of
selecting the right features to enhance model performance, indicating that thorough data preprocessing can greatly
influence the findings from machine learning in the healthcare field.
[15] [25] examined how Convolutional Neural Networks (CNNs) can be applied to determine lung cancer in
radiological pictures for example CT scans and chest X-rays. The research included teaching a model for deep learning
that uses a vast collection of labeled pictorial data, enabling the CNN to grasp complex patterns linked to lung cancer.
The researchers evaluated how well the CNN performed compared to conventional diagnostic techniques,
demonstrating the possiblility of using deep learning to imaging in medicine. With an astounding 97% accuracy rate,
the CNN algorithm far outperformed traditional methods. The writers emphasized that deep learning methods have the
capacity to automate image analysis, lessening the workload for radiologists and enhancing diagnostic precision. This
research represented a major progress in the application of AI in medical imaging, indicating that deep learning has the
capacity to completely transform lung cancer detection.
Liu and associates investigated the efficiency of CNNs and other deep learning models in contrast to conventional
machine learning algorithms aimed at identifying lung cancer in CT scans [16], [17]. The scientists used a dataset
containing different lung cancer cases for a thorough assessment of model performance. Different CNN architectures
were tested to find the best configuration for this particular task, with an emphasis on maximizing both efficiency and
precision of the model. The findings revealedthat CNN models consistently performed better than traditional
algorithms, achieving a 95% correct rate. The research emphasized the benefits of deep learning in capturing intricate
features from imaging data that traditional methods frequently overlook. The researchers determined that deep learning
could greatly improve the ability to diagnose lung cancer in radiology.
Zhang and colleagues carried out a comprehensive analysis of severalmachine learning techniques employing a dataset
of people with lung cancer, including Random Forests, K-Nearest Neighbors (KNN), and Logistic Regression [18],
[19]. The study aimed to identify the best effective method for identifying lung cancer early using clinical and imaging
information. The researchers utilized cross-validation methods to ensure the strength of their results and to prevent
overfitting. The study found that Random Forests reached an accuracy rate of 94%, illustrating a solid combination of
precision and recall. The researchers determined that ensemble proceduresfor example Random Forests are highly
successful in medical diagnosis because of their ability to handle complex datasets andavoid overfitting, rendering them
applicable for practical use in detecting lung cancer.
Table 1 summarizes key studies on lung cancer prediction, highlighting their methodological strengths such as high
accuracy, effective use of CNNs, EHR integration, transfer learning, and radiomics-based approaches. However, it also
notes common weaknesses, including limited datasets, dependency on data quality, lack of interpretability, and
challenges in generalizing results across domains.
Table 1. Strengths and Weaknesses of the current systems
Study Strengths Weaknesses
Ausawalaithong et al. High accuracy (93.5%) with CNNs Limited dataset, black-box
(2019) on X-rays interpretability [20]

Yeh et al. (2020) Uses EHR for early risk prediction Dependent on EHR quality and
(90.2%) completeness [21]

Copyright to IJARSCT DOI: 10.48175/IJARSCT-29107 55

[Link]
 IJARSCT
International Journal of Advanced Research in Science, Communication and Technology
International Open-Access,
Access, Double
Double-Blind, Peer-Reviewed,
Reviewed, Refereed, Multidisciplinary Online Journal

ISSN: 2581-9429 Volume 5, Issue 1, October 2025 Impact Factor: 7.67

Islam et al. (2019) Effective use of transfer learning May not generalize well to new
(95.1%) domains [22]

Li et al. (2020) Combines radiomics and ML Limited by radiomics data extraction [23]
(92.5%)

Wang et al. (2019) CNN on CT scans with strong Dataset size limits scalability [24]
results (94.2%)

These findings underscore the growing success of ML, particularly deep learning, in identifying lung cancer from
complex clinical and imaging data. However, challenges such as interpretability, generalizability, and integration into
clinical workflows remain
in open research questions.

II. METHODOLOGY
A. Introduction
This chapter presents the methodology for creating an artificial intelligence-based
intelligence (ML) lung cancer prediction system.
Emphasis is placed on the design of the dataset pipeline, preprocessing procedures, classifier architecture, performance
evaluation, and tool selection. Ensuring validity and reliability, the section lays the groundwork for replicability and
future extension.

B. Proposed System

Fig. 1. System Architecture of Lung Disease Prediction

A comparative study is conducted using four supervised classifiers:
classifiers Support Vector Machine (SVM)
(SVM), Random
Forest Classifier (RFC), K-Nearest Neighbours (KNN), and Logistic Regression (LR), to forecast lung cancer risk.
Input features include age, smoking habit, air pollution exposure, genetic predisposition, symptoms, and biomarker
indicators. Models are trained to classify patients into risk categories: Low, Medium, or High.
Prior to model training, dataata undergoes comprehensive preprocessing: cleaning, imputing, normalization, feature
selection, and class balancing. Hyperparameter tuning (e.g., SVM kernel, number of trees for RFC, optimal k in KNN)
is performed via grid search and k-fold
fold cross-checking.
cross . Measures of performance include precision, accuracy, recall,
and F1-score.

C. Discussion of Dataset
Dataset Description and Collection
The dataset originates from a publicly available Kaggle repository called "Lung Cancer Risk & Prediction Dataset,"
Dataset,"[29]
which contains approximately 1,000 records and 25 attributes,
attributes including demographics, lifestyle, environmental
exposure, symptoms, and a three-level
level risk target variable (Low, Medium, High).
High)
Features utilized include age, gender, air pollution index, smoking and passive smoking, occupational hazards, genetic
risk, chronic respiratory disease history, and clinical symptoms (chest pain, fatigue, weight loss, etc.). The target

Copyright to IJARSCT DOI: 10.48175/

10.48175/IJARSCT-29107 56
[Link]
 IJARSCT
International Journal of Advanced Research in Science, Communication and Technology
International Open-Access,
Access, Double
Double-Blind, Peer-Reviewed,
Reviewed, Refereed, Multidisciplinary Online Journal

ISSN: 2581-9429 Volume 5, Issue 1, October 2025 Impact Factor: 7.67

variable indicates lung cancer risk level. Dataset labels and features were ext
extracted
racted in a manner consistent with other
published studies.
Table 2: Description of Dataset Features for Lung Cancer Prediction
S/N Feature Description
1 Index Unique ID of records
2 Age Patient’s age
3 Gender Patient’s gender (1 = Male, 2 = Female)
4 Air Pollution Exposure level to air pollutants (scale of 1–10)
5 Alcohol Use Frequency of alcohol consumption (scale of 1–10)
6 Dust Allergy Sensitivity to dust allergens (scale of 1–10)
7 Occupational Hazards Exposure to hazardous work conditions (scale of 1–10)10)
8 Genetic Risk Family history of lung cancer (scale of 1–10)
9 Chronic Lung Disease Presence of chronic lung diseases (scale of 1–10)
10 Balanced Diet Quality of diet (scale of 1–10)
11 Overweightness Obesity level (scale of 1–10)
12 Smoking Smoking frequency (scale of 1–10)
13 Passive Smoker Exposure to secondhand smoke (scale of 1–10)
14 Chest Ache Intensity of chest pain (scale of 1–10)
15 Coughing of Blood Frequency of coughing up blood (scale of 1–10)
16 Exhaustion Level of fatigue (scale of 1–10)
17 Weight Loss Extent of weight loss (scale of 1–10)
18 Shortness of Breath Severity of breathlessness (scale of 1–10)
19 Gasping Intensity of wheezing (scale of 1–10)
20 Swallowing Difficulty Difficulty in swallowing (scale of 1–10)
21 Clubbing of Finger Nails Changes in nail appearance (scale of 1–10)
22 Frequent Cold Frequency of catching colds (scale of 1–10)
23 Dry Cough Severity of dry cough (scale of 1–10)
24 Snoring Frequency of snoring (scale of 1–10)
Target variable (Low, Medium, High – indicating lung cancer
25 Level
risk)

Figure 3 shows the statistical summary of a lung cancer dataset, containing the lowest, quartiles, maximum, count,
mean, and standard deviation values for every aspect.
aspect Figure 4 indicates that the dataset has no missing values, as all
features show a count of zero null entries.

Fig. 3. Describing the dataset

Copyright to IJARSCT DOI: 10.48175/

10.48175/IJARSCT-29107 57
[Link]
 IJARSCT
International Journal of Advanced Research in Science, Communication and Technology
International Open-Access,
Access, Double
Double-Blind, Peer-Reviewed,
Reviewed, Refereed, Multidisciplinary Online Journal

ISSN: 2581-9429 Volume 5, Issue 1, October 2025 Impact Factor: 7.67

D. Data Pre-processing
 Data Cleaning: Missing or invalid entries were handled via imputation or removal as appropriate.
 Standardization: Feature scaling was performed using scikit-learn’s
scikit StandardScaler,, ensuring zero mean and
unit variance.
 Feature Selection: Key predictors were identified using scikit-learn’s
scikit SelectKBest with chi-squared
chi scoring to
reduce dimensionality and
nd enhance interpretability.
 In order to discourse class
lass Imbalance: The Synthetic Minority Over-Sampling Technique (SMOTE) remained
applied via the imbalanced-learn package to balance class distribution and prevent bias toward majority classes
classes.
 Train-Test Split: Using scikit-learn's
learn's train_test_split function, the cleaned dataset was separated into subgroups
for training (80%) and testing (20%).

E. Model Selection and Training

Each algorithm employed is briefly described below and trained on the preprocessed dataset:
1. Support Vector Machine (SVM): Constructs a hyperplane to maximize class separation margin. Kernel and
regularization parameters were tuned.
2. Logistic Regression (LR): Predicts probabilities via the sigmoid function. Implemented using the ‘lbfgs’ solver
so
with regularization parameter C tuning.
3. Using the Manhattan or Euclidean distance metric, K-Nearest
K Neighbors (KNN) is a non-parametric
parametric classifier it
forecasts outcomes by using the majority label among k neighbors. Grid search was used to get the ide
ideal k.
4. Random ForestClassifier (RFC): A bootstrap-aggregated
bootstrap aggregated ensemble of decision trees. In tune among the options
were min_samples_split, max_depth, and multiple estimators (n_estimators).

F. Hyperparameter Tuning
Each machine learning algorithm's hyperparameters
hyperparameters were adjusted to maximize its performance in lung cancer
prediction. The specific hyperparameters tuned for each model included:
1. Support Vector Machine (SVM): In order to maximize the model’s ability to distinguish between lung cancer risk
levels, the regulationconstraint (C) and the kernel type (example.,
( ., linear, polynomial, RBF) were changed.
2. K-Nearest Neighbors (KNN): To increase the precision of categorizing whereas balancing bias and variance, the
number of neighbors (K)K) and the distance metric (such as Manhattan or Euclidean) were adjusted.
3. Random Forest Classifier: To increase model performance, decrease overfitting, and improve generalization in
lung cancer classification, the number of decision trees (n_estimators), maximum tree depth, and minimum
number of samples needed for node splitting were tuned.
Copyright to IJARSCT DOI: 10.48175/
10.48175/IJARSCT-29107 58
[Link]
 IJARSCT
International Journal of Advanced Research in Science, Communication and Technology
International Open-Access,
Access, Double
Double-Blind, Peer-Reviewed,
Reviewed, Refereed, Multidisciplinary Online Journal

ISSN: 2581-9429 Volume 5, Issue 1, October 2025 Impact Factor: 7.67

4. Logistic Regression: To guarantee that the model performed effectively when applied to unknown data, the
regularization parameter (C) was adjusted to prevent overfitting.
overfitt
Hyperparameter tuning's objective was to achieve the highest possible accuracy by tailoring each model to the unique
characteristics of lung cancer risk factors. This approach ensured a well-optimized
well optimized model setup, maximizing predictive
performance in lung cancer classification

G. System Design
Figure 2 outlines the sequential workflow of a disease prediction system, starting from entering patient details to
validating data, extracting features, and matching values. It then proceeds to classify the data, predict the disease, and
finally display the results.

Figure 2. Illustrates the Use Case Diagram, showing user interactions, data input, analytics processing, and output
classification.

III. RESULTS AND FINDINGS

A. Result of Model Training
This project trained four machine learning models: Random Forest Classifier, K-Nearest-Neighbors,
K Neighbors, Support Vector
Machine, and Logistic Regression Classifier. An 80-20
80 20 dataset split was used to train the logistic regression model. The
training data allowed the modelto identify patterns and categorize new occurrencesefficiently. Its prediction skills were
measuredby means of a range of performance pointers. The KNN model was trained using 80% of the dataset, with the
remaining 20% set aside for testing. The classifier
lassifier used 3 nearest neighbors and the Minkowski distance (p=2).
Since KNN is a non-parametric model, instead of learning, commits the training data to memory explicit outlines.
Predictions were made by assigning new data points to their closest neighbor’s predominant class. The SVM model was
trained by means of an 80-20 train-examination
examination split with Bagging to improve generalization. It utilized an RBF kernel
with C=7 and trained 75 SVM classifiers about various data subsets. The One-vs-Rest strategy handled multi-class
classification, and the model's performance was evaluated by means of a confusion matrix and accuracy score on the
test set.

Copyright to IJARSCT DOI: 10.48175/

10.48175/IJARSCT-29107 59
[Link]
 IJARSCT
International Journal of Advanced Research in Science, Communication and Technology
International Open-Access, Double-Blind, Peer-Reviewed, Refereed, Multidisciplinary Online Journal

ISSN: 2581-9429 Volume 5, Issue 1, October 2025 Impact Factor: 7.67

The Random Forest model was trained on a preprocessed dataset, where categorical variables were encoded, and highly
correlated features (Pearson > 0.9) were removed. An 80-20 train-test split was used with stratification to maintain class
balance. To reduce overfitting, the model was fine-tuned by:
1. Limiting tree depth and number of trees
2. Increasing the minimum samples required for splits and leaf nodes
3. Using fewer features per tree and reducing sample size per tree
4. Applying class weighting to handle class imbalance

A. Performance Evaluation
Table 2: Classification Report Logistic Regression
Class Precision Recall F-1 Score Support

1 0.92 0.84 0.88 67

2 0.81 0.88 0.84 58

3 0.95 0.96 0.95 95

Overall Accuracy: Macro Avg: 0.89 Weighted Avg: 0.90

0.90

Logistic Regression had high accuracy and was particularly effective in predicting class 3. However, it showed slightly
lower recall for class 1, which means some high-risk patients may have been misclassified.
K-Nearest Neighbors (KNN): The KNN model attained an accuracy of 0.86, performing best for class 1 with a recall of
0.97, meaning it correctly identified most patients in this category. However, it struggled with class 2, where recall
dropped to 0.69, leading to a higher number of false negatives. The weighted F1-score of 0.86 recommends that while
the model is fairly balanced, its classification of class 2 could be improved. Table 3 presents precision, recall, F1-score,
and support for each class, showing an overall model accuracy of 86%.
Table 3: Classification Report KNN
Class Precision Recall F-1 Score Support

1 0.83 0.97 0.90 67

2 0.89 0.61 0.78 58

3 0.88 0.91 0.89 75

Overall Accuracy: Macro Avg: 0.87 Weighted Avg: 0.86

0.86

KNN was particularly strong in identifying class 1 but struggled with class 2. The high recall for class 1 suggests that
KNN can effectively detect early-stage lung cancer cases but may require tuning for better performance across all
classes.
Support Vector Machine (SVM): SVM demonstrated an accuracy of 0.88, showing balanced performance across all
classes. It performed exceptionally well for class 3, achieving a recall of 1.00, meaning it correctly identified all
patients in this category. However, its performance for class 2 was slightly weaker, with a recall of 0.69. The weighted
F1-score of 0.88 indicates a well-rounded model. Table 4 presents precision, recall, F1-score, and support for each
class, showing an overall model accuracy of 88%.
Copyright to IJARSCT DOI: 10.48175/IJARSCT-29107 60
[Link]
 IJARSCT
International Journal of Advanced Research in Science, Communication and Technology
International Open-Access, Double-Blind, Peer-Reviewed, Refereed, Multidisciplinary Online Journal

ISSN: 2581-9429 Volume 5, Issue 1, October 2025 Impact Factor: 7.67

Table 4: Classification Report SVM

Class Precision Recall F-1 Score Support

1 0.94 0.91 0.92 67

2 1.00 0.69 0.82 58

3 0.79 1.00 0.88 75

Overall Macro Avg: 0.91 Weighted Avg: 0.88

Accuracy:0.88

SVM had a perfect recall for class 3, making it ideal for identifying confirmed lung cancer cases. However, its lower
recall for class 2 suggests potential improvements in fine-tuning the model parameters.
Random Forest Classifier: The Random Forest classifier achieved an accuracy of 0.83. It performed best for class 0,
with a recall of 1.00, meaning all patients in this category were correctly identified. However, its recall for class 2 was
0.53, indicating a high number of false negatives. The weighted F1-score of 0.82 suggests that while the model is strong
in certain areas, it struggles with class 2. Table 5 presents precision, recall, F1-score, and support for each class,
showing an overall model accuracy of 83%.
Table 5: Classification Report Random Forest
Class Precision Recall F-1 Score Support

1 0.84 1.00 0.91 73

2 0.76 0.97 0.85 61

3 1.00 0.53 0.69 66

Overall Accuracy: Macro Avg: 0.87 Weighted Avg: 0.82

0.83

Random Forest was highly effective at predicting class 0 but had lower recall for class 2, meaning it failed to correctly
identify a significant number of patients in that category. This suggests that feature selection or additional data
balancing techniques may improve its performance.
In summary, Among the models, Logistic Regression achieved the highest accuracy (90%), demonstrating strong
overall performance across all lung cancer risk levels. It was particularly effective in predicting class 3 cases, with a
high recall and F1-score, making it a reliable choice for identifying lung cancer patients.
SVM performed exceptionally well in detecting advanced lung cancer cases (class 3), achieving a recall of 1.00 for this
category. However, it showed lower recall for class 2, indicating that some cases were misclassified.
KNN was most effective in identifying early-stage lung cancer cases (class 1), with a recall of 0.97. However, its
performance dropped for class 2, leading to a higher number of false negatives in that category.
Random Forest exhibited strong predictive capabilities for class 0, achieving a perfect recall of 1.00. However, it
struggled with class 2, where recall was only 0.53, suggesting a need for further tuning or feature selection to improve
its performance.
Overall, Logistic Regression emerged as the best-performing model due to its balanced precision, recall, and high
accuracy. SVM and KNN also showed strong predictive abilities, particularly for specific lung cancer stages. Random
Forest, while robust, would benefit from hyperparameter tuning and data balancing techniques to enhance its
classification performance.
Copyright to IJARSCT DOI: 10.48175/IJARSCT-29107 61
[Link]
 IJARSCT
International Journal of Advanced Research in Science, Communication and Technology
International Open-Access, Double-Blind, Peer-Reviewed, Refereed, Multidisciplinary Online Journal

ISSN: 2581-9429 Volume 5, Issue 1, October 2025 Impact Factor: 7.67

IV. SUMMARY AND CONCLUSION

A. Summary
The paper examined and contrasted the efficiency of four machine learning algorithms in forecasting the peril of lung
cancer: Support Vector Machines (SVM), Random Forest Classifier, K-Nearest Neighbours (KNN), and Logistic
Regression. The study used a Kaggle dataset and adhered to a methodical approach include acquiring data,
preprocessing, choosing features, training the model, adjusting hyperparameters, and assessing performance based on
accuracy, precision, recall, and F1-score metrics.
Among the models assessed, Logistic Regression demonstrated the highest accuracy and overall balanced performance,
turning it into the most actual algorithm for lung cancer prediction in this study. Each algorithm showed distinct
strengths and weaknesses, providing valuable insights into their practical applicability for early lung cancer detection.
The paper highlights the promising part of machine learning techniques in supporting clinical decision-making and
enhancing diagnostic accuracy for lung cancer patients.

B. Limitations
Despite the encouraging results, this study has limitations. First, the dataset was sourced from Kaggle and may not
represent the full spectrum of lung cancer cases across diverse demographics and geographic regions, potentially
limiting the generalizability of the findings. Second, the study relied on a limited set of features; important clinical or
genetic factors not included could influence model accuracy and robustness. These constraints suggest that real-world
application requires further validation and broader feature integration.

C. Future Work
Future research could improve lung cancer prediction by exploring sophisticated methods like deep learning with
Artificial Neural Networks (ANNs), which may better capture complex medical data patterns. Expanding datasets to
include more diverse populations and additional clinical and genetic features would enhance model generalizability and
accuracy. Additionally, including these integrating clinical decision support systems with predictive models with
predictive models might enable real-time diagnosis, enabling earlier intervention and improved patient outcomes.

D. Conclusion
In summary, this research shows how machine learning may help with the early diagnosis of lung cancer using
accessible medical parameters. Logistic Regression emerged as the best-performing model, highlighting its suitability
for classification tasks in lung cancer prediction. While limitations exist, the findings underscore the transformative role
artificial intelligence can play in healthcare, providing a foundation for future development of robust, clinically
applicable diagnostic tools. With continued research and refinement, machine learning models have the capacity to
significantly enhance lung cancer detection and patient care.

REFERENCES
[1] Cleveland Clinic, “Lungs: Location, Anatomy, Function & Complications,” Cleveland Clinic — Health Library.
[Online]. Available: [Link] (Verified)
[2] American Cancer Society, “Key Statistics for Lung Cancer,” 2022. [Online]. Available:
[Link] (Verified)
[3] A. L. Samuel, “Some Studies in Machine Learning Using the Game of Checkers,” IBM Journal of Research and
Development, vol. 3, no. 3, pp. 210–229, 1959.
[4] E. E. Onuiri, O. Ukandu, and K. Umeaka, “International Journal of Research Publication and Reviews Machine
Learning Models for Lung Cancer Subtype Classification: A Systematic Review,” International Journal of Research
Publication and Reviews, vol. 5, no. 9, pp. 1299–1308, 2024.
[5] T. M. Cover and P. E. Hart, “Nearest Neighbor Pattern Classification,” IEEE Transactions on Information Theory,
vol. 13, no. 1, pp. 21–27, Jan. 1967, doi: 10.1109/TIT.1967.1053964.
[6] C. Cortes and V. Vapnik, “Support-Vector Networks,” Machine Learning, vol. 20, no. 3, pp. 273–297, 1995.
Copyright to IJARSCT DOI: 10.48175/IJARSCT-29107 62
[Link]
 IJARSCT
International Journal of Advanced Research in Science, Communication and Technology
International Open-Access, Double-Blind, Peer-Reviewed, Refereed, Multidisciplinary Online Journal

ISSN: 2581-9429 Volume 5, Issue 1, October 2025 Impact Factor: 7.67

[7] L. Breiman, “Random Forests,” Machine Learning, vol. 45, no. 1, pp. 5–32, 2001. [Online]. Available:
[Link] (Verified)
[8] D. W. Hosmer and S. Lemeshow, Applied Logistic Regression, 2nd ed. New York: Wiley, 2000.
[9] S. Garg, P. Pundir, G. Rathee, P. K. Gupta, S. Garg, and S. Ahlawat, “On Continuous Integration / Continuous
Delivery for Automated Deployment of Machine Learning Models using MLOps,” arXiv, preprint arXiv:2202.03541,
Feb. 2022. [Online]. Available: [Link] (URL added)
[10] W. D. Travis, E. Brambilla, A. P. Burke, A. Marx, and A. G. Nicholson, “The 2015 World Health Organization
Classification of Lung Tumors,” Journal of Thoracic Oncology, vol. 10, no. 9, pp. 1243–1260, Sep. 2015.
[11] National Cancer Institute, “Lung Cancer Prevention,” PDQ® Cancer Information Summaries. [Online]. Available:
[Link] Accessed: Aug. 9, 2025. (Verified)
[12] G. Shmueli and O. R. Koppius, “Predictive Analytics in Information Systems Research,” MIS Quarterly, vol. 35,
no. 3, pp. 553–572, 2011.
[13] H. Sutton, “Peter Morgan Sutton,” BMJ, vol. 348, no. mar31 11, pp. g2466-g2466, Mar. 2014, doi:
10.1136/bmj.g2466.
[14] A. Kumar, S. Singh, and P. Arora, “Comparative Analysis of Machine Learning Algorithms for Lung Cancer
Detection,” Procedia Computer Science, vol. 132, pp. 556–563, 2018.
[15] G. Cai et al., “Medical AI for Early Detection of Lung Cancer: A Survey,” arXiv, preprint arXiv:2410.14769, Oct.
2024. [Online]. Available: [Link] (URL added)
[16] A. Esteva et al., “Dermatologist-Level Classification of Skin Cancer with Deep Neural Networks,” Nature, vol.
542, no. 7639, pp. 115–118, Feb. 2017, doi: 10.1038/nature21056.
[17] Y. Zhang, L. Jiang, and W. Li, “Evaluation of Machine Learning Methods for Lung Cancer Detection Using
Clinical Data,” IEEE Access, vol. 7, pp. 109960–109968, 2019.
[18] A. B. Goldberg, “SSL with Realistic Tuning,” University of Wisconsin–Madison, Computer Sciences, Tech. Rep.,
2009.
[19] A. Chaudhari, A. Singh, S. Gajbhiye, and P. Agrawal, “Lung Cancer Detection using Deep Learning,” arXiv,
preprint arXiv:2501.07197, Jan. 2025. [Online]. Available: [Link] (URL added)
[20] W. Ausawalaithong, S. Marukatat, A. Thirach, and T. Wilaiprasitporn, “Automatic Lung Cancer Prediction from
Chest X-ray Images Using Deep Learning Approach,” arXiv, preprint arXiv:1808.10858, Aug. 2018. [Online].
Available: [Link] (URL added)
[21] M. C. H. Yeh et al., “Artificial Intelligence–Based Prediction of Lung Cancer Risk Using Nonimaging Electronic
Medical Records: Deep Learning Approach,” Journal of Medical Internet Research, vol. 23, no. 8, e26256, Aug. 2021,
doi: 10.2196/26256.
[22] M. I. Islam et al., “VER-Net: a hybrid transfer learning model for lung cancer detection using CT scan images,”
BMC Medical Imaging, vol. 24, Art. no. 98, May 2024.
[23] J. Li, Z. Li, L. Wei, and X. Zhang, “Machine Learning in Lung Cancer Radiomics,” Machine Intelligence
Research, vol. 20, no. 6, pp. 753–782, 2023, doi: 10.1007/s11633-022-1364-x.
[24] J. Wang et al., “Lung Cancer Detection Using Co-learning from Chest CT Images and Clinical Demographics,”
arXiv, preprint arXiv:1902.08236, Feb. 2019. [Online]. Available: [Link] (URL added)
[25] A. Kumar, S. Singh, and P. Arora, “Comparative Analysis of Machine Learning Algorithms for Lung Cancer
Detection,” Procedia Computer Science, vol. 132, pp. 556–563, 2018. (Same as [14])
[26] T. M. Cover and P. E. Hart, “Nearest Neighbor Pattern Classification,” IEEE Transactions on Information Theory,
vol. 13, no. 1, pp. 21–27, Jan. 1967, doi: 10.1109/TIT.1967.1053964. (Same as [5])
[27] T. Hastie, R. Tibshirani, and J. Friedman, The Elements of Statistical Learning, 2nd ed. Springer, 2009.
[28] A. B. Goldberg, “SSL with Realistic Tuning,” University of Wisconsin–Madison, Computer Sciences, Tech. Rep.,
2009. (Same as [18])
[29] "Lung Cancer Risk & Prediction Dataset," Kaggle, Accessed: Sep. 18, 2025. [Online]. Available:
[Link]

Copyright to IJARSCT DOI: 10.48175/IJARSCT-29107 63

[Link]