Machine Learning Framework for Urine Screening and Early Disease Detection in Women’s Health

Non-invasive screening can improve early disease detection and reduce the need for invasive diagnostics. This study introduces a machine learning framework designed for multipathology screening based on urinary fluorescence data. A dataset of urine samples from women across several diagnostic categories was analyzed using standardized preprocessing and feature engineering procedures. Multiple machine learning models, including Logistic Regression, Support Vector Machine, Random Forest, Extreme Gradient Boosting, and Stochastic Gradient Descent, were trained and validated using repeated stratified cross-validation. All classifiers demonstrated strong and consistent performance, with accuracy values above 0.80 and area-under-the-curve scores between 0.85 and 0.90. Model interpretability analysis identified stable diagnostic features contributing to discrimination between healthy and pathological samples. The proposed workflow offers a fast, low-cost, and scalable solution for early disease screening.