A novel Bayesian Pay-As-You-Drive insurance model with risk prediction and causal mapping

The modern vehicle insurance industry is increasingly adopting Pay-As-You-Drive (PAYD) insurance models, aligning premium costs with driving behavior. Our study introduces a Bayesian approach to PAYD insurance, leveraging the strengths of Naive Bayes classifiers and Bayesian Networks to handle uncertainty and integrate prior knowledge in risk assessment. The Naive Bayes model achieved an 87.5% accuracy in predicting risk partitions. With the Bayesian Network providing insights into causal relationships through a Directed Acyclic Graph (DAG), we also address the challenges of traditional actuarial models — low interpretability of intra-factor relationships and thus hard to plan for risk management for both provider and policyholder. Our research contributes to optimizing insurance pricing strategies. Still, the causal mapping also dismisses the meaningfulness of using geographic grouping in insurance pricing (discriminatory or not). It reassures the theoretical advantage of the PAYD model over the traditional model, facilitating access to affordable coverage for policyholders. • Use Bayesian models to Pay-As-You-Drive insurance for more accurate risk assessment. • Train Naive Bayes that achieves 87.5% accuracy in predicting risk. • Apply domain expertise to provide “prior” in Bayesian Network causal modeling. • Utilize DAGs to model causal links between risk factors and insurance claims. • Build dynamic causal maps of driving behaviors to guide interventions.