Multi-view ensemble learning for predictive analytics of telematics data using Bayesian model averaging

Abstract Telematics has emerged as one of the most critical data sources in today’s auto insurance industry. Despite the potential predictive value of telematics information for auto insurance claims, such data are typically available for only a small subset of policyholders, leaving the majority without telematics features. In this paper, we propose a multi-view ensemble learning framework to address this unique data structure. We consider multiple predictive models trained on different data views and combine these heterogeneous models through Bayesian model averaging. Through an extensive experimental study, we demonstrate that the proposed method assigns data-adaptive weights to individual models and delivers robust predictive performance across a variety of practical settings.