D. Anaya Luque, L. Bermúdez Morata, J. Belles Sampera
Explainable artificial intelligence (xAI) aims to enhance our understanding of the decision-making processes and outcomes produced by opaque machine learning (ML) models. In this context, we present various xAI techniques that can empower risk managers with more transparent ML methods. We demonstrate this through an application focused on improving the management of paid-up risk in insurance savings products. To achieve this, we utilize a database of actual universal life policies to develop an initial logistic regression model as well as multiple tree-based models. Subsequently, we employ different xAI techniques, in conjunction with the utilization of a Kohonen network, to provide valuable insights into the functioning of the tree-based models for end-users. Drawing from these findings, we showcase how novel and non-trivial concepts can emerge to enhance paid-up risk management.
Palabras clave: Machine learning, Shapley values, Kohonen networks, Risk analysis
Programado
GT02.AR1 Análisis Riesgo
7 de noviembre de 2023 18:40
HC3: Sala Canónigos 3