Evolutionary Hyperparameter Optimization of Machine Learning Classifiers for Gallstone Disease Prediction

Abstract

This study investigate whether evolutionary hyperparameter search can meaningfully improve classifier performance for gallstone disease (GSD) diagnosis. Working from a cohort of 319 patients described by 38 clinical and demographic variables, we split the data into training (80%) and testing (20%) partitions using stratified sampling, and standardized features using statistics computed solely from the training partition to avoid information leakage. A Genetic Algorithm (GA) then searched the hyperparameter space of six classifiers– Random Forest (RF), Extreme Gradient Boosting (XGBoost), Support Vector Machine (SVM), Decision Tree (DT), K-Nearest Neighbors (KNN), and Logistic Regression (LR) with each candidate configuration scored by 5-fold stratified cross-validation restricted to the training partition. Tuned models were refit on the full training set and evaluated once, on the held-out test partition, using accuracy, precision, recall, F1-score, and confusion-matrix analysis. XGBoost led every metric at 90.62%, with a ROC-AUC of 93.46%, RF and SVM tied at 79.69% accuracy, LR reached 78.12%, KNN 67.19%, and DT62.50%. The margin between XGBoost and the remaining classifiers indicates that, for this dataset, pairing gradient boosting with evolutionary tuning offers a genuine advantage over the alternatives tested.