Clinical decision system for chronic kidney disease staging using machine learning

Abstract

Background:

Chronic Kidney Disease (CKD) is a prevalent health condition that requires personalized treatment planning at each of its five stages. Machine Learning (ML) and Generative AI have shown promise in predicting CKD progression based on patient data. However, existing prediction models have limitations on generalizability, interpretability, and resource requirements.

Objective:

This study aims to develop a clinical support system using ML models to classify CKD stages accurately. The research focuses on feature selection strategies and model performance evaluation to enhance prediction accuracy and guide personalized treatment planning for CKD patients.

Methods:

The study utilizes ML algorithms, including Gradient Boosting, XGBoost, CatBoost, and GAN AML, to categorize CKD stages. Various feature selection techniques such as Recursive Feature Elimination, chi-square test, and SHAP are employed to identify relevant features for improved prediction accuracy. The models are evaluated based on precision, recall, F1-score, accuracy, and AUC-ROC metrics.

Conclusions:

The findings demonstrate the effectiveness of CatBoost and GAN AML in accurately classifying CKD stages, highlighting the importance of expert knowledge in selecting feature selection strategies to enhance ML model performance. Future research directions include validating diverse datasets, integrating with clinical practice, and improving interpretability and explainability in CKD prediction models.