Kidney Stone Prediction Based on Urine Analysis Using Ensemble Learning

Kidney stone formation is a prevalent and painful condition affecting millions worldwide. This work describes a machine learning approach for predicting the onset of nephrolithiasis by analyzing urine samples collected over a 24-hour period. The objective is to develop a consistent and effective method for early identification and prevention in order to lessen the need for intrusive and costly therapies. Our research capitalizes on ensemble learning methodologies, encompassing Random Forest, XGBoost, AdaBoost, and LightGBM, to scrutinize urinary parameters and ascertain individuals predisposed to renal calculi prior to the onset of stone formation. The study utilizes a dataset from Kaggle containing 7 features and 769 samples. Feature extraction and selection techniques were employed to enhance model performance. With an accuracy of 98.48%, precision of 98.90%, recall of 97.83%, and F1-score of 98.36%, the Random Forest classifier performed the best. What turned out to be the most informative predictors were gravity, calcium content, and their ratio. Kidney stone illness may be less common, healthcare expenses may be decreased, and patient outcomes may be enhanced by this non-invasive, economical method.