Predicting overactive bladder symptom severity with explainable AI: a data-driven approach using questionnaire responses

Abstract

Overactive bladder (OAB) significantly affects an individual’s quality of life by disrupting daily routines and social interactions. The hallmark symptom of OAB is a sudden, uncontrollable urge to urinate, often resulting in involuntary leakage. Diagnosing OAB is challenging due to the subjective nature of symptom reporting and the absence of definitive biomarkers. Existing diagnostic methods, such as urodynamic testing and symptom questionnaires, provide valuable insights but often lack conclusive reliability. This research leverages a data-driven approach to diagnose and predict the severity of OAB symptoms using comprehensive questionnaire data. Participants of varying ages provided demographic information, medical history, and details such as urination frequency. OAB severity was classified into three levels: no OAB, mild OAB, and moderate OAB, reflecting an increasing intensity of symptoms. The collected data were used to train multiple machine learning (ML) models, including support vector machine (SVM), artificial neural networks (ANN), linear discriminant analysis (LDA), and random undersampling boosted (RUSBoost). Model performance was evaluated using K-fold cross-validation (K = 5 and 10), measuring accuracy, recall, and F1 score to ensure generalizability. Among the models, SVM achieved the highest accuracy, with a 10-fold cross-validation accuracy of 93.33%. To address limitations such as small dataset size and class imbalance, the synthetic minority oversampling technique (SMOTE) was applied, further enhancing model performance. Explainable artificial intelligence (XAI) techniques were also implemented, showing how individual features influenced model predictions. This eliminated the need for manual intervention by uncovering intricate patterns in symptom data and making the diagnostic process more accessible and interpretable. This study underscores the potential of integrating machine learning into OAB diagnosis. The results demonstrate that questionnaire-based predictions of OAB severity are highly accurate and cost-effective, surpassing the performance of human experts. This approach offers a promising solution for enhancing patient care and streamlining OAB management by reducing diagnostic costs and improving clinical decision-making.