Machine Learning-Driven Visualization for Early CKD Prediction in Smart Health

This study proposes a clinical visualization method for evaluating machine learning algorithms in early chronic kidney disease (CKD) detection. By integrating k-nearest neighbors (KNN), support vector machines (SVM), and logistic regression (LR) with 3D biomarker mapping techniques, we validate the approach through two clinical scenarios: (1) diabetes risk stratification (based on blood glucose, age, and hemoglobin) and (2) hypertension-kidney interactions (based on blood pressure, sodium, and potassium). In the diabetes analysis, the KNN algorithm preserves gradual progression patterns of biomarkers, while SVM generates abrupt classification boundaries conflicting with current clinical guidelines. For the hypertension scenario, KNN demonstrates spatial consistency with anatomically distributed renal function deterioration cases, whereas LR exhibits ambiguous prediction confidence near critical clinical thresholds. The visualization method enables direct comparison between the predictions and CKD staging criteria through juxtaposed displays, visually revealing spatial correspondences between machine learning outputs and raw biomarker distributions. This work provides medical researchers with an interpretable paradigm for clinical implementation of machine learning.