InsurFraud is an interdisciplinary research project at the crossroad of Economics/Econometrics and Machine Learning (ML), which aims to contribute to the development and dissemination (in the academic and professional fields) of recent advances in Data Science applied to the field of insurance fraud detection. Importantly, this fraud is not a recent phenomenon, but it still continues to have huge financial, economic and social impacts. Nowadays, insurance fraud records impressive amounts and it still shows an up-trend due to the expansion of modern technologies. Within the InsurFraud project, our goal is to identify the main built-in domain specificities of insurance fraud and to propose appropriate solutions to detect it. Moreover, the partnership with a private insurance company allows us to use high-quality proprietary data sets and valuable expert knowledge on insurance fraud cases.

The project is designed around four main objectives. First, fraud is a compex, time-evolving and intentional phenomenon. Insurance fraud can take a wide variety of forms (e.g., automobile insurance fraud, household insurance fraud, etc.), and even though their characteristics are the same and the final objective consists in detecting the fraud cases, the solution to each problem is rather domain-specific. Additionally, fraudsters have a highly adaptive behavior over time and their actions are intentional. For all these reasons, the first objective of the project is to develop combined, automated and real-time detection methods to spot them. 

Second, fraud is a rare-event, meaning that the datasets used for fraud detection are highly imbalanced, which makes the detection process even more complicated. The class imbalance issue heavily compromises both the process of learning and the evaluation of its accuracy. Our objective is to assess the behavior of different methodologies when dealing with imbalanced data, both theoretically and empirically.

Third, the evaluation of fraud detection models is crucial and raises various problems. InsurFraud will focus predominantly on the construction of new statistical tests used to assess the performance of different classifiers.

Finally, the fourth objective of InsurFraud project is to conduct an academic research analysis on the trade-off between the statistical performance of the predictive models, the economic cost of fraud and the limited financial and human resources used for the investigation of alerts. 

[More details HERE]