Predicting Lung Cancer Using Clinical Data and Machine Learning

Abstract

Early and accurate lung cancer detection remains a critical challenge in medical diagnostics. This study investigates the application of multiple machine learning classifiers combined with data resampling and hyperparameter optimization techniques for improved predictive performance on lung cancer datasets. Synthetic Minority Oversampling Technique (SMOTE) was employed to resolve severe class imbalance, while Jellyfish Search Optimizer (JSO), a metaheuristic algorithm, was utilized to optimize hyperparameters, particularly of the Random Forest classifier. Experimental results indicate that integrating SMOTE and JSO enhances model generalizability and predictive accuracy, with the Random Forest classifier demonstrating the highest improvement accuracy increasing from 91.94% to 95.16% and F1 score from 95% to 97%. Other classifiers such as AdaBoost and Decision Trees experienced minimal performance changes, signifying that ensemble methods combined with optimization are promising for lung cancer prediction tasks. Multiple metrics including Accuracy, F1 score, ROC_AUC, and PR_AUC were used to provide comprehensive evaluation, underscoring the robustness of the proposed methodology. The findings suggest potential applications in clinical decision support systems.