Whitening-based Adversarial Bi-Regressor for Cross-platform Insurance Claims Prediction

Insurance claims prediction aims to predict claim amounts for auto-adjusting premiums. Deep learning has shown great potential in this field but typically requires large amounts of labeled data. In our cross-platform insurance claims business, obtaining labeled data is highly challenging due to privacy and confidentiality concerns. Our business focuses on providing trained models to partners (other insurance companies) without accessing their data labels. This creates a challenge: applying models trained on our private data directly to partners' data. This problem falls under Unsupervised Domain Adaptation (UDA). Additionally, since we aim to predict claim amounts, it specifically belongs to the Unsupervised Domain Adaptation Regression (UDAR) task.