Predict fraudulent transactions in the cashiers of our stores with the company owned data about our transactions.
We mostly consider internal frauds, which are more difficult to detect. The cases listed below may contain fraudulent activities:
- Customers purchasing the same products from the same cashier.
- Hidden gross sales, in which customers abuse the company policies, possibly by dividing their purchases into smaller quantities.
- Product returns, such as products returned with a different price.
- Basket related fraud, such as a customer basket that is not coherent with other baskets or customers abusing self-service registers.
The challenge aims to detect scenarios or transactions related to internal fraud. We provide a set of validated fraudulent transactions. These transactions constitute labelled data for training the fraud detection algorithms and we use existing cash register video analysis systems to correlate project findings with real transactions.
The challenge has the following sample datasets available for download
- Transaction header information
- Transaction items
- Customer purchase history
- Anonymised employee information
The proposed software should be used to:
- To provide accurate predictions to prevent fraud.
- To provide decision makers insight for considering other fraud prevention measures, such as installing camera surveillance systems for stores.
We expect the proposed solution to:
- Be scalable enough to be used in a major chain of stores (as it is Migros).
- Provide insight into what factors can help us determine a fraudulent transaction.
- Provide accurate predictions in terms of true positives and true negatives.
- Reduce loses by 10%.