Machine Learning-Based Classification of Lower Urinary Tract Symptoms from Uroflowmetry: Evaluation Against Physician Decisions and Clinical Nomograms

Lower Urinary Tract Symptoms (LUTS) are commonly evaluated using clinician judgment and nomogram-based methods derived from limited physiological parameters. However, these approaches may not generalize across patient subgroups and may not fully capture patient-specific variability. we develop Machine Learning (ML) models to predict the necessity of medical intervention using uroflowmetry and clinical features. Unlike prior work focusing on symptom clustering or parameter-based analysis, the proposed approach models physician-assessed decisions and enables a direct comparison with established clinical nomograms. Multiple ML models, including Support Vector Machines (SVM), Random Forest (RF), and gradient boosting methods, as well as neural network architectures, are evaluated on a clinically collected dataset using stratified analysis across age and sex groups. Experimental results show that ML-based models achieve higher overall classification accuracy than Liverpool and Siroky nomograms, particularly in male subgroups. Reduced performance in female subgroups highlights the impact of data imbalance and suggests the need for more representative datasets. These findings suggest that ML-based approaches can provide more flexible and data-driven decision support compared to nomogram methods.