An Investigation Into Modeling Non-Life Insurance Claims In The Nigerian And Ghanaian Insurance Market

Abstract

For calculating non-life insurance premiums, actuaries rely on separate modeling of frequency and severity using covariates to explain the claims loss exposure. In this thesis, we focus on the insurance claims severity amount. Two separate insurance claims data were analyzed using some selected Tree-Based Machine Learning (ML) Algorithms namely; the Classification and Regression Tree (CART), Random Forest (RF), and Gradient Boosting (GB) Models. The predictive performance of the selected models were compared using the Coefficient of determination (R2), Mean Absolute Error (MAE), and Mean Squared Error (MSE). In the application of the selected models, this Thesis relied on two different insurance claims data; The Nigerian and the Ghanaian Insurance claims dataset. The Nigerian dataset had 10,017 observations from paid claims with 4 explanatory variables, while the Ghanaian dataset had 5,495 observations with 7 explanatory variables. In the analysis, 70% of the data were used for training and 30% for testing. Both datasets were compared in terms of the selected performance measures. The results show that the Random Forest model of the claims amount had the overall best performance for both Nigerian and Ghanaian Dataset.