Classifying chronic kidney disease using selected machine learning techniques

Chronic kidney disease (CKD) is a serious global health problem with high mortality rates, often due to late diagnosis. Early detection and classification are essential to improve treatment outcomes and slow disease progression. This study evaluates the performance of four machine learning algorithms—linear discriminant analysis (LDA), Naïve Bayes, C4.5 decision tree, and Random Forest—in classifying CKD using a Kaggle dataset containing 1,659 instances and 52 features, covering demographic, lifestyle, and clinical data. After data pre-processing, the classification accuracies of the algorithms were assessed. LDA showed the highest accuracy at 92.8%, followed by Naïve Bayes (92.1%), C4.5 (92.0%), and Random Forest (91.9%) before hyperparameter tuning. After tuning, C4.5 achieved the highest accuracy of 92.5%, followed by Random Forest (92.2%), with Naïve Bayes remaining at 92.1%. However, even after tuning, LDA remained the most accurate, demonstrating superior performance. The key features contributing to CKD classification were serum creatinine, glomerular filtration rate (GFR), muscle cramps, protein in urine, fasting blood sugar, itching, systolic blood pressure, blood urea nitrogen (BUN), HbA1c, edema, total cholesterol, body mass index (BMI), and gender. These findings confirm that LDA outperforms other algorithms in CKD classification without the need for tuning, emphasizing the value of machine learning in improving early diagnosis and management of CKD.