Explainable AI for Kidney Stone Prediction: A Machine Learning-Based Clinical Analysis

This study aims to apply machine learning (ML) methods to accurately predict kidney stones by utilizing essential clinical parameters. Kidney stones are a prevalent and painful urological disorder that has repercussions on patient health and healthcare expenditure. Tried and true diagnostics methods, such as imaging, are often expensive and require time. A robust predictive framework has been developed in this study based on a dataset that has six essential clinical features (gravity, pH, osmolality, conductivity, urea, and calcium). As preparation for effective modeling, multiple preprocessing techniques like outlier detection, feature scaling, and balancing of data with the help of SMOTE (Synthetic Minority Over-sampling Technique) were applied. We trained and optimized multiple machine learning algorithms using hyperparameter tuning, including Support Vector Machines (SVM), K-Nearest Neighbors (KNN), Random Forest, and Extra Trees Classifier. Out of all models, the Extra Trees Classifier has shown to be the most effective, registering an accuracy of 92.5% with a high ROC-AUC of 0.96, which indicates excellent discriminatory ability. To improve the interpretability of the predictions and shed light on important factors leading to predictions, interpretability techniques were utilized including SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations) to identify drivers in the formation of kidney stones. The strongest predictors identified were calcium, osmolality, and urea. In the future, we plan to include additional clinical and genetic information to improve the proposed system’s predictive capability and real-world applicability.