Intelligent Recommendation Framework for Optimizing Risk in Medical Health Insurance Using Clustering and Association Rule Mining

Abstract

The medical insurance industry is confronted with rising claim volatility and increasingly diverse customer profiles. Traditional actuarial techniques often struggle to capture subtle interactions among demographic, behavioral and clinical features, leading to coarse premium segmentation and sub‑optimal risk sharing. This paper proposes an intelligent recommendation framework to optimize risk in medical health insurance. The approach uses a real claims dataset sourced from Kaggle, containing demographic, occupational and hereditary attributes. After cleaning duplicates, replacing missing values and applying Min–Max scaling and label encoding, exploratory analysis visualizes disease count distributions, hereditary conditions and job‑title correlations. K‑means clustering is employed to discover natural risk segments; optimum cluster numbers are determined using the silhouette and elbow methods. Cluster assignments are then used as input for association rule mining with Apriori and FP‑Growth algorithms. Support, confidence and lift measures guide rule selection. The system ultimately produces rule‑based recommendations for new applicants, predicting their likely claim group and suggesting premium strategies. Experiments demonstrate that seven clusters achieve a high silhouette score while seven clusters minimize within‑cluster variation. FP‑Growth is found to be faster and generates more concise rules than Apriori. The proposed framework aids insurance providers in understanding risk patterns, designing personalized products and reducing adverse selection. Future work will investigate deep learning and federated techniques to enhance scalability and privacy.