Gallstone Disease Prediction Using Clinical and Biochemical Features Through Ensemble Learning Techniques

Abstract

This paper presents an ensemble learning-based approach for predicting gallstone disease using clinical and biochemical indicators. With the increasing prevalence of gallstones due to aging populations and evolving lifestyles, early diagnosis is essential for timely and effective treatment. Gallstones are typically classified into cholesterol, pigment, mixed, and rare types, and their chemical composition plays a critical role in determining appropriate medical interventions. This study focuses on leveraging both categorical and numerical patient data to improve predictive accuracy. A voting-based ensemble framework was developed by integrating multiple classifiers, including Random Forest, XGBoost, Logistic Regression, Gradient Boosting, Extra Trees, LightGBM, and Support Vector Machine. The model was trained and evaluated on a dataset containing individual clinical and biochemical records. Initial results showed a baseline accuracy of 79.69% and an area under the ROC curve of 0.9238-the highest achieved in the series of experiments. After applying optimal thresholding (threshold = 0.40), the accuracy increased to 87.50%, with the model maintaining a high recall of 94% for positive cases. The proposed method not only enhances diagnostic accuracy through the power of ensemble techniques but also improves model transparency by linking SHAP-ranked features directly to medically relevant factors. These findings confirm that the ensemble approach outperforms standalone models and provides a reliable, data-driven support system for clinicians. Beyond accurate prediction, this method contributes to the broader vision of developing smart healthcare systems by integrating predictive modeling into clinical workflows.