Revolutionizing Early Lung Cancer Detection with AI and ML Solutions

Abstract

Lung cancer is the leading cause of cancer-related deaths worldwide, and early detection is critical for improving patient survival. Advances in machine learning (ML) provide powerful tools for analyzing medical data, yet it is unclear which ML approaches are most effective for lung cancer prediction. This study tested the hypothesis that ensemble methods combining multiple ML models would achieve higher predictive accuracy than individual models, reaching over 90% accuracy for both patient demographic models and convolutional neural network (CNN) models. It was further predicted that increasing hyperparameters (such as tree depth, number of estimators, or epochs) would improve accuracy but reduce computational speed. A publicly available dataset from Kaggle containing clinical and diagnostic data was used. Four models were trained and evaluated: Decision Tree, Random Forest, Logistic Regression, and a Classifier algorithm. Their outputs were then combined using the Multiplicative Weight Update Method to create an ensemble prediction. Model performance was evaluated using accuracy, precision, and recall. Results showed that the Decision Tree achieved 92.2% accuracy, Random Forest 95.1%, Logistic Regression 65.1%, and the Classifier 91.2%. The ensemble model significantly improved prediction, reaching 96.62% accuracy. Hyperparameter tuning further improved accuracy but at the cost of slower performance. These findings support the hypothesis and highlight the trade-off between accuracy and computational efficiency in ML-based lung cancer prediction.