AI-Based Early Detection and Risk Prediction of Jaundice Using Clinical and Liver Function Test Data

Abstract: Jaundice, the yellow discolouration of skin and sclera caused by elevated serum bilirubin, is a clinically visible manifestation of hepatic, haematological or biliary dysfunction whose early detection materially affects patient outcomes. In the Indian subcontinent, where viral hepatitis A and E are endemic and non-alcoholic fatty liver disease is rising in prevalence, the burden of liver disease is substantial while specialist hepatology expertise is concentrated in tier-one urban centres. This paper presents HepatIQ, an artificial-intelligence-driven decision support system for the early detection and risk stratification of jaundice. The system combines a Random Forest classifier trained on the Indian Liver Patient Dataset of 583 records with a rule-based pattern classifier and a biochemical flagging engine to produce explainable risk assessments. The system achieves a five-fold cross-validation accuracy of 70.68% (±3.14%) and a test- set accuracy of 72.65%, with a clinically critical recall of 100% on the high-risk class at 97% precision. The hybrid combination of probabilistic machine learning output, pattern classification, biochemical flags and India-specific dietary recommendations addresses the explainability gap that limits adoption of black-box predictors in clinical settings. The complete system is delivered as a Flask-based web application with SQLite persistence, runs on commodity hardware, and uses only open-source libraries.

Keywords: Jaundice, Liver Function Test, Random Forest, Machine Learning, Indian Liver Patient Dataset, Risk Stratification, Clinical Decision Support, Explainable AI, Hepatology, Bilirubin.