Advancing Insurance Intelligence: Integrated Statistical and Machine Learning Models in Loss Reserving and Auto Insurance Claim Prediction using Telematics data

Abstract

This thesis marks a significant stride forward in insurance modeling, emphasizing innovative methodologies for loss reserving and auto insurance claim prediction utilizing telematics data. Spanning three distinct research avenues, it tackles the complexities of modeling loss reserve data and harnesses telematics data to enable accurate risk assessment in non-life insurance settings.

In the first research avenue, Bayesian loss reserve models are introduced to model loss reserve data within run-off triangles. These models incorporate persistence terms within the conditional autoregressive range model to effectively account for claims persistence over time. To enhance flexibility, various error distributions are explored, with models featuring log-transformed mean functions and persistence terms demonstrating superior fits.

Transitioning to the second research avenue, a comprehensive analysis of a large telematics dataset is conducted to predict future claims based on driving behavior. The objective is to classify drivers into safe and risky groups and derive premiums that accurately reflect actual driving risk. Two-stage Poisson, Poisson mixture, and zero-inflated Poisson regression models with lasso regularization are employed.

In the third research avenue, the exploration of neural networks in insurance claim prediction introduces the Poisson mixture deep learning neural network. Through meticulous search techniques optimizing network architecture, notable improvements in prediction accuracy are achieved, paving the way for enhanced claim prediction capabilities.

In conclusion, this thesis advocates for potential refinements in loss reserves models and proposes the adoption of a 3- or more group Poisson mixture model for analyzing claim counts. These methodological advancements hold significant promise for improving risk assessment and premium pricing strategies in the non-life insurance sector, ultimately benefiting both insurers and policyholders alike.