Scholarlink

Chronic Kidney Disease (CKD) is a significant global health concern, requiring early and accurate diagnosis to prevent progression to end-stage renal disease. This study presents a machine learning-based approach for CKD prediction using Random Forest (RF) and XGBoost (XGB) classifiers. The research emphasizes the integration of feature selection and SHAP (SHapley Additive exPlanations) analysis to optimize performance and enhance interpretability. Mutual Information was used as the feature selection method to identify the most relevant features from a dataset of 400 samples. The optimized XGBoost model, using an 11-feature subset, achieved a maximum accuracy of 99.17%, while the RF model attained 98.33% accuracy. SHAP values were used to interpret the impact of individual features, revealing that specific gravity, serum creatinine, blood urea, and hemoglobin levels were the most critical contributors to the prediction outcomes. These results demonstrate the effectiveness of combining feature selection and explainable AI techniques in building high-performance CKD diagnostic models. The findings were compared with previous studies, indicating superior performance and practical implications for clinical decision-making. This study provides a robust, interpretable model for CKD prediction that could support healthcare professionals in early diagnosis and personalized patient management.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing Chronic Kidney Disease Prediction Using Explainable Machine Learning Models

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Prokash Gogoi,
  • J. Arul Valan

摘要

Chronic Kidney Disease (CKD) is a significant global health concern, requiring early and accurate diagnosis to prevent progression to end-stage renal disease. This study presents a machine learning-based approach for CKD prediction using Random Forest (RF) and XGBoost (XGB) classifiers. The research emphasizes the integration of feature selection and SHAP (SHapley Additive exPlanations) analysis to optimize performance and enhance interpretability. Mutual Information was used as the feature selection method to identify the most relevant features from a dataset of 400 samples. The optimized XGBoost model, using an 11-feature subset, achieved a maximum accuracy of 99.17%, while the RF model attained 98.33% accuracy. SHAP values were used to interpret the impact of individual features, revealing that specific gravity, serum creatinine, blood urea, and hemoglobin levels were the most critical contributors to the prediction outcomes. These results demonstrate the effectiveness of combining feature selection and explainable AI techniques in building high-performance CKD diagnostic models. The findings were compared with previous studies, indicating superior performance and practical implications for clinical decision-making. This study provides a robust, interpretable model for CKD prediction that could support healthcare professionals in early diagnosis and personalized patient management.