Gallstone disease classification using SLOA-optimized CatBoost classifier with explainable AI.

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Data Science for Disease Prediction

Background:

• Gallstones affect many individuals, often asymptomatically.
• While image-based machine learning (ML) is established for gallstone detection, research on tabular data is less explored.
• This study addresses the gap by analyzing tabular data for gallstone prediction.

Purpose of the Study:

• To investigate predictive models for gallstone presence using a publicly available tabular dataset.
• To compare the performance of standard Catboost (CB) with Sea Lion Optimization Algorithm (SLOA)-optimized CB.
• To identify key features influencing gallstone prediction through model explainability techniques.

Main Methods:

• Utilized the Catboost (CB) classifier with 5-fold cross-validation.
• Employed the Sea Lion Optimization Algorithm (SLOA) to optimize the CB model, including feature selection.
• Applied SHAP, LIME, and DiCE for model explainability analysis.

Main Results:

• The standard CB model achieved a mean accuracy of 79.58% with 38 features.
• The SLOA-optimized CB model reached a mean accuracy of 80.42% using 19 selected features.
• C-Reactive Protein (CRP) and Vitamin D were identified as the most influential features for gallstone prediction.

Conclusions:

• Machine learning models, particularly optimized Catboost, demonstrate effectiveness in predicting gallstones from tabular data.
• Feature selection through optimization algorithms can enhance model performance.
• CRP and Vitamin D levels are significant indicators for gallstone disease prediction, offering potential for early detection strategies.