An Attention-Driven Deep Learning Model for Chronic Kidney Disease Prediction

Abstract

Early detection of chronic kidney disease (CKD) presents challenges, and a late diagnosis may result in significant health complications, including kidney failure and cardiovascular issues. This project sought to develop an AI-driven system utilizing a Transformer model to assess the risk and stage of chronic kidney disease (CKD), thereby aiding physicians in making timely and precise decisions. Patient data, including vital signs, blood and urine test results, and medical history, were gathered and prepared through normalization of numerical features, addressing missing values, and identifying the most significant indicators. The Transformer model was designed to identify patterns within patient records and subsequently evaluated across multiple datasets. It achieved an impressive performance with around 99.2% accuracy, 98% precision, and 97% recall in detecting at-risk patients and accurately categorizing disease stages. The model improved its predictive reliability by focusing on key elements and minimizing the impact of irrelevant information. This AI-driven system could improve patient outcomes, promote preventive healthcare, and provide an effective approach for the early detection of CKD. Furthermore, it could be extended into an easy-to-use continuous monitoring application, allowing physicians and patients to manage kidney health better and reduce the risk of complications through timely interventions.