Dependable AI Machine Learning Models for the Prediction of Urinary System Diseases

Different urinary tract conditions often exhibit overlapping symptoms such as a strong urge to urinate, burning sensations, abnormal urine output and fever, making diagnosis and treatment challenging and time-consuming. Traditional diagnostic methods are slow, leading to delays in treatment. Additionally, a significant issue with AI models is their black-box nature. Thus, this study developed dependable AI machine learning models to predict two urinary system diseases: acute nephritis of the renal pelvis (ANRP) and inflammation of the urinary bladder (IUB). This paper compared twelve machine learning techniques. To address the data imbalance the support vector machine synthetic minority over-sampling technique (SVMSMOTE) is employed. Shapley Additive Explanations (SHAP) values are used to address the black-box nature for interpretability. SHAP tools are employed to demystify the models, providing transparency and identifying key features influencing predictions. This approach ensures the ethical adoption in healthcare by addressing transparency, casuality and interpretability in diagnosis. The results demonstrate performance of 100% across accuracy, precision, recall, and fl- score for SVM, KNN, voting and stacking model, with KNN excelling in training time. SHAP analysis provided valuable insights, making our models clinically relevant.