Comparing Polynomial Logistic Regression and XGBoost for Structured Lung Cancer Risk Prediction Using Demographic, Behavioral, and Symptom-Based Variables

Abstract

The Lung cancer remains one of the leading causes of cancer-related mortality worldwide, and early risk prediction may support timely screening and intervention. This study compared an interpretable regression-based model with a more flexible machine learning model for structured lung cancer prediction. The publicly available datasets was used, which contain demographic, behavioral, and symptom-related variables. After the cleaning and preprocessing, this study developed and evaluated three models, baseline logistic regression model, polynomial logistic regression model, and XGBoost. The polynomial logistic regression model used interaction-only second-order features to capture relationships between predictors. The model performance is evaluated using accuracy, precision, recall, F1-score, ROC-AUC, brier score, and 5-fold cross-calibration. On the hand-out test, the polynomial logistic regression model gets the best overall performance and the XGBoost achieve the highest recall which indicates the stronger sensitivity for detecting positive lung cancer cases. All in all, the results suggest that polynomial logistic regression model provides a strong balance between predictive performance and interpretability, while XGBoost serves as a useful complementary model when sensitivity is prioritized.