ANOVA–Mutual Information Feature Selection with CatBoost for Gallstone Disease Classification

Abstract

Gallstone disease is a common biliary disorder forwhich non-invasive, ultrasonography-independent risk assessmentmay support clinical decision-making. This study proposesan ANOVA–Mutual Information (ANOVA–MI) feature-selectionframework combined with CatBoost, evaluated on 319 patientsand 38 candidate predictors from the UCI Gallstone dataset. Theframework fuses min–max normalized ANOVA F-statistics andmutual-information scores through a tunable weight α, ranksfeatures, and selects compact subsets inside a nested repeatedstratified cross-validation protocol (5×5 outer folds, 3-fold innerOptuna tuning) to prevent information leakage. At α = 0.3,the method retains 15 features (60.5% dimensionality reduction)and achieves an F1-score of 0.800 ± 0.021 and ROC-AUC of0.861±0.032, slightly outperforming the full 38-feature CatBoostbaseline as well as ANOVA-only and MI-only selection. Thecorresponding sensitivity and specificity are 78.2% and 82.6%,respectively. Compared with three recent studies on the samedataset, the proposed approach attains the highest recall andF1-score while using fewer or a comparable number of features.SHAP analysis identifies C-reactive protein and vitamin D as thedominant predictors. These results demonstrate that a modelagnostichybrid of statistical discrimination and informationtheoreticrelevance can substantially reduce feature dimensionalitywhile preserving competitive predictive performance underrigorous nested evaluation.