ML-CKDP: Machine learning-based chronic kidney disease prediction with smart web application.

Area of Science:

• Nephrology and Medical Informatics
• Application of Artificial Intelligence in Healthcare
• Data Science for Disease Prediction

Background:

• Chronic kidney diseases (CKDs) pose significant health risks, necessitating early diagnosis for effective management.
• Machine learning (ML) offers powerful tools for enhancing predictive diagnostics in healthcare.
• Current diagnostic approaches can be improved with advanced computational methods.

Purpose of the Study:

• To develop a machine learning-based kidney disease prediction (ML-CKDP) model.
• To optimize dataset preprocessing for improved CKD classification.
• To create a user-friendly, web-based application for real-time CKD prediction.

Main Methods:

• Implemented a comprehensive preprocessing protocol including data imputation and Min-Max scaling.
• Utilized diverse feature selection techniques (e.g., Correlation, Chi-Square, RFE, Lasso, Ridge).
• Evaluated seven classifiers (RF, AdaB, GB, XgB, NB, SVM, DT) using accuracy, confusion matrix, and AUC metrics.

Main Results:

• Random Forest (RF) and AdaBoost (AdaB) achieved 100% accuracy and AUC across various validation splits (70:30, 80:20, K-Fold 10/15).
• Naive Bayes (NB) demonstrated superior efficiency with the lowest training and testing times.
• A functional web-based application for real-time CKD prediction was successfully deployed.

Conclusions:

• The ML-CKDP model, particularly RF and AdaB, demonstrates exceptional performance in predicting CKDs.
• Feature selection and robust preprocessing are critical for high accuracy in predictive models.
• The developed web application enhances accessibility and practical utility of ML-driven CKD diagnostics.