Nephrosense AI: Risk Detection and Analysis for Chronic Kidney Disease

Chronic kidney disease (CKD) is a condition involving the progressive decline of renal function, which significantly deteriorates a person’s health and quality of life. CKD is characterised by its asymptomatic nature until it reaches an advanced stage, where holistic treatment options become less effective. Traditional diagnostic approaches rely on singular markers; however, these conventional metrics fail to capture the dynamic nature of CKD progression, which is influenced not only by biochemical parameters but also by lifestyle behaviours and genetic predispositions. To address these limitations, there is an urgent need for a comprehensive framework that integrates predictive analytics with patient-centric care. This study proposes Nephrosense AI - a framework designed to predict CKD risk, classify its stages, and generate adaptive treatment plans for patients. The framework utilises a dataset of 4,000 patient records sourced from Kaggle, chosen for its richness in clinical and behavioural features, which are necessary for comprehensive CKD modelling. Nephrosense AI integrates supervised, unsupervised, and reinforcement learning techniques within a modular pipeline to deliver patient-centric insights. It utilises supervised learning for risk prediction and stage classification, unsupervised clustering for patient stratification, and reinforcement learning for adaptive treatment optimisation. The supervised learning module achieved a classification accuracy of 92% and an F1-score of 0.90, with a regression R² of 0.82 for CKD onset prediction - supporting timely diagnosis and more accurate stage identification. The unsupervised clustering module (K-means) produced seven distinct patient clusters with a silhouette score of 0.56, outperforming DBSCAN in clinical relevance, and enabling improved patient stratification for targeted intervention. The reinforcement learning module demonstrated a 15% improvement in patient glomerular filtration rate outcomes after 100 training iterations using Q-learning, directly contributing to enhanced treatment plan refining. Visualisation tools, e.g., dashboards and charts, are integrated with the framework to aid patients in tracking their disease progression and gathering insights. A React-based website with Flask as the backend has been developed to accept user input and provide real-time interaction with the framework. The impact of combining supervised, unsupervised, and reinforcement learning techniques in Nephrosense AI demonstrates that artificial intelligence can significantly enhance CKD diagnosis and management by offering a multi-modal, practical, and adaptive solution, thereby improving patient outcomes and enabling more proactive, data-driven CKD care.