Real-Time Insurance Risk Predication With Continuous Learning and Explainable AI

Insurance underwriting requires accurate and timely assessment of risk to ensure fair premium pricing and sustainable operations for insurance providers. Traditional actuarial models primarily rely on static demographic tables and historical averages, which often fail to capture complex behavioural patterns and nonlinear relationships present in modern healthcare and lifestyle data. To address these limitations, this study proposes AIRIS-X (Autonomous Insurance Risk Intelligence System), a real-time insurance risk prediction framework that integrates machine learning, continuous learning mechanisms, and explainable artificial intelligence. The system collects policyholder data including age, gender, body mass index, number of children, smoking habits, geographic region, pre-existing medical conditions, preferred hospital networks, and co-payment preferences. Using these inputs, a trained XGBoost classifier predicts the probability of insurance risk and categorizes users into low, medium, or highrisk groups. The system addresses dataset imbalance using the Synthetic Minority Oversampling Technique (SMOTE), improving model performance and fairness. Predictions are stored in a centralized database and appended to a continuously growing dataset that allows administrators to retrain the model periodically, enabling adaptive learning from new user inputs. To enhance transparency, SHAP (SHapley Additive exPlanations) is integrated to provide feature-level explanations for each prediction, allowing both insurers and users to understand key risk factors. Additionally, a rule-based recommendation module suggests suitable insurance policies and preventive health measures. A chatbot interface provides contextual assistance and answers user queries related to insurance risk factors. The proposed platform demonstrates an end-to-end intelligent insurance analytics solution that combines predictive modelling, explainability, and continuous data-driven improvement to support modern insurance decision-making systems.