Application of Machine Learning in Actuarial Pricing for Nigerian Motor Insurance

This study examines the application of machine learning techniques in actuarial pricing models for motor insurance in Nigeria, assessing their potential to improve risk classification accuracy relative to traditional generalized linear model approaches. Machine learning methods including gradient boosting, random forest, and neural networks have transformed insurance pricing in advanced markets by enabling more complex risk variable interactions and non-linear relationships to be captured. In Nigeria's motor insurance market, poor risk segmentation is a key driver of adverse selection and market unprofitability. This study uses a motor insurance claims dataset from three cooperating insurers comprising 45,000 policy records from 2020 to 2023, covering vehicles in Lagos, Abuja, Kano, and Port Harcourt. Claims frequency and severity models are built using generalized linear models as a baseline and compared against gradient boosting machine and random forest models. Predictive performance is assessed using Gini coefficient, mean absolute error, and out-of-sample log-likelihood criteria. Findings reveal that gradient boosting models outperform GLMs by 18 percent on the Gini coefficient metric for claims frequency prediction. Vehicle age, driver age, usage type, and geographic zone emerge as the most predictive variables across all models. The study concludes that machine learning significantly improves pricing accuracy for Nigerian motor insurance. It recommends that NAICOM develop a regulatory framework for algorithmic pricing transparency that ensures ML-based pricing models are explainable and non-discriminatory.

Keywords: machine learning, actuarial pricing, motor insurance, gradient boosting, risk classification.