Data Science and Fraud Detection in Insurance

Scientific project

Automatic fraud detection is a very specific domain of statistical modeling. Unlike anomaly detection in the industrial domain for example, it aims at detecting fraudulent transactions resulting from rational agent behavior. Therefore, the detection of fraud cases requires not only skills in data science and econometrics (knowledge of statistical models and their properties), but also economic and legal skills to understand the motivations and strategic behaviors of fraudsters. The motivations of the “Data Science and Fraud Detection in Insurance” IoR are mainly related to (i) the modeling of the strategic behavior of fraudsters in the insurance domain; (ii) the need to set up efficient detection systems given the huge financial losses linked to fraud; (iii) the exploitation of new databases allowing to identify the mechanisms of insurance fraud.

The project is organized around several actions:

Objective:

Enhancing fraud detection in insurance through data science tools while considering the strategic, economic, and legal aspects of fraudulent behaviors.

Research Areas:

1. Detection Methodologies:

   • Development of algorithms specifically designed for fraud detection in insurance.
   • Optimization of the processing of suspicious cases (positive signals) for better resource management.

2. Analysis of Fraudulent Behaviors:

   • Modeling fraudsters’ strategies and the mechanisms behind fraud.
   • Establishing a typology of fraud and fraudsters in non-life insurance.

3. Links to Other Financial Risks:

   • Studying the connections between fraud detection and other risk areas such as credit risk.

4. Governance and Stability:

   • Considering the robustness and management of detection models in an evolving context.

Ambitions:

• Reducing financial losses associated with insurance fraud.
• Training data science experts specialized in fraud detection.
• Disseminating knowledge and methodologies through academic publications, seminars/conferences, and academic training.

Training:

Summer schools and training sessions on machine learning and natural language processing have been organized for researchers and industry professionals.

Scientific Output & Journals:

The project primarily explores financial fraud detection in the insurance sector. It also addresses issues related to social contribution fraud, imbalanced datasets, and the financial and economic evaluation of fraud. These studies combine econometrics, machine learning, and economic approaches. For a complete and updated list of publications and research, please visit the project’s website.

Outreach:

The project promotes its research through conferences, project calls, scientific articles, and training events (summer schools, ad hoc training sessions).

Scientific Lead:

Denisa Banulescu-Radu
Laboratoire d’Économie d’Orléans

Partners:

• Academic: Laboratoire d’Économie d’Orléans
• Economic: Thélem Assurances

Summary:

The “Data Science and Fraud Detection in Insurance” initiative brings together experts in econometrics, data science, and economics to develop innovative techniques for preventing and detecting insurance fraud. A recruited PhD student explores these complex issues, integrating economic and data science dimensions. The project tackles several major challenges: modeling fraudsters’ strategic behavior, managing imbalanced datasets, and developing technical tools for real-time fraud detection. Dissemination of results is supported through conferences, publications, and training sessions.

Website:

Data Science and Fraud Detection in Insurance Project