University of Nairobi Team: Felix Omondi, Allan Waithaka, Esther Auma, Patricia Teresa and Stacy Jebet
This project explored how data and statistical modelling can help insurance companies better understand and predict motor insurance claim amounts. Real insurance data is often, skewed, and far from “normal,”. As such, traditional linear models don’t always work well. To address this, the project used Generalized Linear Models (GLMs) — a flexible modelling approach widely used in modern actuarial work.
Modelling Process
The study used a real motor‑insurance dataset with over 9,000 customer records, including details such as income, education, vehicle type, premium paid, gender, and how long the customer had been with the insurer. Two types of GLMs were built and compared:
- A Gamma model, and
- An Inverse Gaussian model
The project followed a full modelling process:
From the results, the model showed that Education, Employment Status, Marital Status, Vehicle Size and Vehicle Class were the significant variables in explaining the claim amounts.
The final model was then used to predict future claim amounts and assess the effectiveness of GLMs in providing reliable claim estimates.
Beyond motor insurance, the project highlighted how predictive modelling supports pricing, customer insights, and risk assessment — all essential parts of today’s data‑driven insurance industry.
Read the full research paper below: