Asia Pacific University Library catalogue

A thesis submitted in fulfillment of the requirement for the award of the degree of Master of Science in Data Science and Business Analytics (UCMF1808DSBA)

In the era of big data expansion, how to use data to improve risk assessment is a key point for insurance companies. The underwriting process is the starting step of an insurance policy and the first customer touch point. The life insurance underwriters currently face the problem of how to find a solution to improve the accuracy of risk assessment and service efficiency at the same time. This article proposes a solution with three research objectives for underwriters to overcome this predicament. As the high dimensional dataset became the common challenge for insurance dataset, the first objective would be demonstrating the impact between different dimension reduction techniques. One filter method and one wrapper method of feature selection will be applied in this research. The second objective would be identifying the most key risk factors for risk assessment in underwriting, in order to improve the quality of data collection for better risk management. And the last objective is comparing the performance between different machine learning algorithms. In this research, Multiple Linear Regression (MLR), XBoost, Support Vector Regression (SVR) and Stacking Ensemble model be trained according to those two feature selection methods. As the results, overall the models built based on wrapper method have the better performance, meanwhile, Stacking Ensemble model achieved the best performance with RMSE as 1.92 and MAE as 1.45, respectively. Furthermore, this study also analysed the most significant factors that influence the risk level most according t the feature selection methods and models.