A text analytics-based decision support system for detecting fraudulent car insurance claims

Bravo, Cristian, Medina, Andrés, Joannon, Rodrigo and Weber, Richard (2018) A text analytics-based decision support system for detecting fraudulent car insurance claims. 29th European Conference on Operational Research, , Valencia, Spain. 08 - 11 Jul 2018.

Abstract

Car insurance is a highly competitive business line in the insurance
industry. Companies face a very large number of claims year-on-year
and must decide carefully when to doubt a given claim, or when to simply cover the accident without question. In this presentation, we will show the results of a decision support system that ranks claims by their level of risk, suggesting which claims should be checked by an expert agent from the insurance agency. The decision support system analyses both common structured data, such as claim times and claimant information, and the report made by the claimant in free-text format. We show different strategies to deal with the text information, from using simple bag-of-words models, to much more complex embeddings using transforms such as Latent Semantic Analysis and fastText embeddings. Both sources of data, text-based and structured, are then used to create different predictive models estimating the probability that any given claim is fraudulent. Results indicate that there are clear gains in using more sophisticated models, but that data quality, specially having reliable fraud labels, might force the use of simpler models. We also will discuss how these models change business practices, by streamlining fraud detection as part of the day-to-day operation of insurance companies.

This record has no associated files available for download.

Contributors

Author: Cristian Bravo

Download statistics