Development and optimization of predictive machine learning models for patients with chronic kidney disease

Abstract

Introduction. Artificial intelligence has evolved, becoming an essential tool for analyzing information in medicine. Objective. Develop, optimize, and compare the performance of different machine learning models for analyzing factors associated with chronic kidney disease. Methodology. Machine learning models were developed to predict chronic kidney disease. Logistic regression models, support vector machines, random forests, and decision trees were constructed, and performance metrics were evaluated. The model with the lowest performance was selected and optimized using conventional machine learning techniques, hyperparameter tuning, and advanced approaches. Performance was evaluated using accuracy, area under the curve, sensitivity, specificity, 95% confidence intervals, and p-values < 0.05. Results. Logistic regression stood out for its accuracy (85.29%) and sensitivity (95.65%), and support vector machines for the area under the curve (92.09%). Random forest achieved a balance between accuracy (82.35%) and area under the curve (90.32%). The tree showed high specificity (90.91%) and positive predictive value (90%). After hyperparameter tuning, the decision tree achieved an accuracy of 80.39%. Conclusion. Logistic regression, support vector machines, and random forest performed best with conventional training. Machine learning techniques allowed the performance of the models to be adjusted and optimized, and male sex, high blood pressure, and exposure to pesticides were identified as key factors associated with of chronic kidney disease.