Abstract:
Accurate claims reserving is an important aspect for all insurers in the insurance business to meet their future legal liabilities. Despite many methods being proposed for prediction of claim reserves,
it still remains a challenge in the industry and in particular, Malawian insurance industry is no exception. In recent years Machine Learning (ML) has caught many attentions across many industries, specifically ML has a vast range of applications in the insurance industry. This research aimed at investigating the use case of ML approaches namely; Random Forest (RF), Extreme Gra dient Boosting (XGB) and Neural Networks (NN), on individual claim reserve predictions in auto insurance. Using scikit-learn and keras in Python, Random Forest, Extreme Gradient Boosting and Neural Network regressions were used on Malawian insurance claims data. Prior to model fitting,the data was exposed to various ways of preprocessing techniques such data cleaning, data trans formation, feature selection and feature engineering where one-hot-encoding was used to transform
all categorical variables into numerics. Then the data was splitted into two sets; one set for model training (80%) and the other for model validation (20%). In order to select the model that pro vides the most accurate claim reserve amounts in the auto insurance business, various evaluation criteria including Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE) and accuracy score (R2) were used. The results showed that RF achieved an accuracy score of 89%, MAE of 2.78, and RMSE of 3.91 while XGB and NN both achieved an accuracy score of 88%. Further, XGB achieved a MAE of 2.92 and RMSE of 4.11 while NN achieve a MAE of 2.91 and RMSE of 4.13. Thus, overall all the three models have produced desirable results as they have all achieved high accuracy scores. However, RF has slightly outperformed the other two methods and
hence may be regarded as the best model out of the three. This implies that ML regression models, in particular RF, XGB and NN have the ability of producing the most accurate individual claim reserve predictions in auto insurance industry and they may therefore be considered in prediction of claims reserves.
