A Improving lung cancer classification with feature selection: a comparative study of random forest and xgboost

Abstract

The leading cause of cancer mortality worldwide remains lung cancer which can be better managed when early and precise diagnosis is achieved to enhance patient outcomes. High-dimensional datasets in medical diagnostics create obstacles for classification because redundant and irrelevant features diminish model accuracy and boost computational complexity. This research investigates how feature selection enhances the performance of lung cancer classification models. The study evaluates Random Forest (RF) and XGBoost as classification models and uses Genetic Algorithm (GA) for feature selection to enhance model efficiency. The GA process ran for 50 generations and reached convergence at the 40th generation which showed that the optimal feature subset had reached stability. Random Forest outperformed XGBoost using GA-based feature selection in a number of parameters, such as accuracy, precision, recall, F1-score, and AUC-ROC. Random Forest displays superior effectiveness in utilizing optimized feature subsets to achieve enhanced generalization and classification performance over XGBoost. The research stands out because it compares how feature selection affects RF and XGBoost algorithms for lung cancer classification using fixed model settings. The research findings demonstrate the value of integrating RF with GA for feature selection as it offers potential for building both efficient and interpretable lung cancer diagnostic models within medical AI