Metaheuristic-based gallstone classification using rotational forest explained with SHAP.

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Digestive Disease Research

Background:

• Gallstone disease (cholelithiasis) affects millions globally, presenting diagnostic challenges due to variable symptoms and potential severe complications like pancreatitis and gallbladder cancer.
• Early prediction of gallstones is crucial for timely medical intervention and improved patient outcomes.
• Existing diagnostic methods may not fully capture the complexity of gallstone disease risk factors.

Purpose of the Study:

• To develop and evaluate a novel machine learning approach for the early prediction of gallstones.
• To optimize a Rotational Forest (RoF) classifier using the Bald Eagle Search (BES) algorithm for enhanced gallstone prediction accuracy.
• To identify the most significant predictive features for gallstone formation.

Main Methods:

• A tabular dataset was utilized to train and test a Rotational Forest (RoF) classifier.
• The RoF model was optimized using the Bald Eagle Search (BES) algorithm to improve predictive performance.
• Feature importance was analyzed using SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME).

Main Results:

• The RoF classifier alone achieved 78% accuracy and an AUC of 0.867 using all features.
• The RoF model optimized with BES achieved 75.78% accuracy and an AUC of 0.860, utilizing a reduced feature set of 17.
• Key predictive features identified include C-reactive protein (CRP), Vitamin D levels, Obesity, Hemoglobin (HGB), and Body Mass Index (BMI).

Conclusions:

• The optimized RoF model demonstrates effective gallstone prediction capabilities.
• Feature importance analysis highlights key biomarkers and clinical factors associated with gallstone disease.
• This approach offers a promising tool for early gallstone detection and risk stratification.