Machine Learning Driven Prediction of Chronic Nephrology Related Diseases

Chronic Kidney Disease (CKD) presents a significant public health challenge due to its increasing prevalence and life-threatening consequences. With an average survival time of only 18 days without functioning kidneys, the urgency for kidney transplants and dialysis treatments underscores the importance of early prediction. This research proposes a machine learning-driven approach for CKD prediction using clinical data. The workflow includes advanced preprocessing techniques, such as collaborative filtering for missing value imputation and meticulous attribute selection, to optimize model performance. Among the 11 machine learning methods evaluated, XGBoost and transformer models demonstrated superior accuracy, reducing bias towards attributes and ensuring reliable predictions. The study emphasizes the crucial role of practical data collection and domain expertise in fine-tuning machine learning models for effective CKD detection. By achieving high accuracy, this research aims to enhance early CKD detection and management, contributing to improved healthcare outcomes and reduced burdens on medical systems.