Automatic analysis of insurance reports through deep neural networks to identify severe claims

Abstract In this paper, we develop a methodology to automatically classify claims using the information contained in text reports (redacted at their opening). From automatic analysis, aim is predict if claim expected be particularly severe or not. The difficulty rarity of such extreme database, and hence difficulty, for classical prediction techniques like logistic regression accurately outcome. Since data unbalanced (too few observations are associated with positive label), propose different rebalance algorithm deal issue. We discuss use embedding methodologies used process data, role architectures networks.