Leveraging xAI for enhanced surrender risk management in life insurance products

Abstract

Explainable Artificial Intelligence (xAI) plays a crucial role in enhancing our understanding of decision-making processes within black-box Machine Learning models. Our objective is to introduce various xAI methodologies, providing risk managers with accessible approaches to model interpretation. To exemplify this, we present a case study focused on mitigating surrender risk in insurance savings products. We begin by using real data from universal life policies to build logistic regression and tree-based models. Using a range of xAI techniques, we gain valuable insight into the inner workings of tree-based models. We then propose a novel supervised clustering approach that integrates Shapley values with a Kohonen neural network (KNN). The process involves three main steps: computing Shapley values from a supervised tree-based model; clustering individuals into homogeneous profiles using an unsupervised KNN; and interpreting these profiles with a supervised decision tree model. Finally, we present several key findings derived from the application of xAI techniques, which have the potential to enhance surrender risk management practices.