Optimizing Claims Risk Prediction for Commercial Health Insurance Users by Combining the DeepFM Algorithm
F. Mi, S. H. Mi · Advanced Electromagnetics · 2026
Commercial health insurance claims prediction is challenged by label delay, temporal distribution drift, and complex high-dimensional feature interactions, which reduce the reliability and stability of conventional risk assessment models. To address these issues, this study proposes a DeepFM-based dual-correction framework that integrates inverse probability weighting, symmetric cross-entropy loss, temporal adversarial learning, and maximum mean discrepancy constraints to jointly mitigate label lag and distribution drift. A shared embedding architecture is employed to capture both low-order and high-order feature interactions, while an adaptive gated fusion strategy balances the contributions of the FM and DNN branches for robust probability estimation. Experimental results demonstrate that the proposed model achieves an AUC of 0.853 under standard conditions and maintains competitive performance under severe label delay and temporal drift scenarios, with limited degradation in calibration metrics. The framework significantly enhances prediction accuracy, robustness, and probability consistency for commercial health insurance claims risk assessment. Furthermore, its capability for modeling dynamic feature interactions and time-varying data distributions provides methodological insights for intelligent signal interpretation and adaptive prediction tasks in electromagnetic wave propagation environments and related antenna data analysis applications.