Gallstone Classification Using Random Forest Optimized by Sand Cat Swarm Optimization Algorithm with SHAP and...

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Computational Biology

Background:

• Gallstone disease impacts 10-20% of adults globally, necessitating early diagnosis.
• Machine learning (ML) shows promise for gallstone detection, yet tabular data approaches are underexplored.
• Optimizing diagnostic models with clinical data is crucial for effective gallstone management.

Purpose of the Study:

• To develop and evaluate a Random Forest (RF) classifier optimized with Sand Cat Swarm Optimization (SCSO) for gallstone prediction using tabular clinical data.
• To assess the impact of cross-validation and feature reduction on model performance.
• To identify key predictive features and enhance model interpretability.

Main Methods:

• A Random Forest (RF) classifier was developed and optimized using the Sand Cat Swarm Optimization (SCSO) algorithm.
• Experiments were conducted using four frameworks: RF without cross-validation (CV), RF with CV, RF-SCSO without CV, and RF-SCSO with CV.
• Feature importance was analyzed using SHAP, and model interpretability was enhanced with DiCE.

Main Results:

• The RF-SCSO model achieved comparable accuracy to the standard RF model while significantly reducing the feature set from 38 to 13.
• SHAP analysis identified C-reactive protein (CRP), Vitamin D, and Aspartate Aminotransferase (AAST) as highly influential features.
• DiCE analysis provided corrective counterfactuals for misclassified instances, enhancing model transparency.

Conclusions:

• Optimized ML models, particularly RF-SCSO, offer a viable approach for gallstone diagnosis using structured clinical data.
• Feature selection and interpretability methods like SHAP and DiCE are valuable for refining diagnostic ML models.
• This study highlights the potential of tabular data and interpretable ML in improving gallstone disease management.