Lung Cancer Stage Detection using Tree-Based Ensemble Algorithms: A Benchmarking Study with Feature Selection and ROC Analysis

The world-wide leading cause of cancer-related deaths due to lung cancer needs early stage detection which leads to better patient results. The research examines the outcome of ensemble methods Random Forest and Gradient Boosting alongside XGBoost and LightGBM in lung cancer stage identification by analyzing structured medical data. The analysis used systematic methods which first cleaned data before choosing features through Random Forest algorithms and then employing PCA for dimension reduction. A complete assessment of the severe models took place through examining combination diagnostic metrics consisting of accuracy together with AUC-ROC and precision-recall curves as well as confusion matrices. The XGBoost model emerged as the top ensemble model with 90.1% predictive accuracy and AUC-ROC score of 0.950 while Gradient Boosting delivered 87.8% accuracy alongside 0.930 AUC before Random Forest reached 86.7% accuracy and 0.910 AUC then LightGBM reached 75.0% accuracy with 0.850 AUC score. The research outcome showed Coughing of Blood in combination with Passive Smoking and Obesity as main indicators for predicting the outcome. The developed models demonstrated robust behavior using visual insights that showed ROC curves alongside heatmaps alongside learning curves to establish interpretation capabilities of these models. Research outcomes demonstrate that ensemble methods particularly boosting approaches generate superior stepwise lung cancer diagnostic performance regarding accuracy and dependability. Active patient care will benefit from the system through implementing three development stages that unite genomic data with imaging information and optimize model parameters.