Healthcare Fraud Detection at a bank

Development of a supervised machine learning model to identify healthcare fraud


Every year investigative journalists for FTM (Follow The Money) publish a list of fraudulent healthcare companies based on their annual financial reports. Margins above 10% and high dividend payments are typical indicators of fraud. However analysing financial statements can be very time consuming and often these can be redacted in order to hide the fraudulent behaviour. It is hence preferable to find a way to automatically monitor the transactions of these companies and detect fraud without delay.


In collaboration with the client, RiskQuest devised a model that combines simple business rules with a supervised ML (machine-learning) algorithm based on random forests. The model was developed in Python by leveraging the computational power of an Azure Databricks cluster. The team first established a set of business rules that in each month identified the customers that would have to be monitored. These rules were created by combining the vision of stakeholders with data analytics. By cleverly selecting the appropriate features, a supervised random forest model was then trained based on monthly aggregated transactional data.


The model was approved by the client’s Model Validation department and later put in production. By combining statistical techniques and business knowledge the team was able to successfully come up with a method which can detect healthcare fraud. Illicit activities can be spotted in a short timeframe by monitoring a company’s transactional behaviour. Such a model can be efficiently maintained throughout the years and also serves as a general framework for other types of fraud detection efforts. RiskQuest was part of each step of the modelling process, as well as the stakeholder management with compliance, legal and management departments. Our team's knowledge of the financial crime domain as well as both quantitative and technical experience were instrumental in creating intelligent features to identify this type of fraud.

Related cases
Anti-money laundering
Related cases
Automatic retraining and deployment
Related cases
Improve sensor activity for a care institution