Financial Fraud Alert Review Dataset

The FiFAR dataset, is comprised of 30K bank account opening application instances, accompanied by the judgments of a team of 50 synthetic fraud analysts with realistic decision-making properties on whether or not each instance is a fraudulent application. Each instance contains information regarding the bank account opening application and the applicant, as well as the ground truth label: 0 - legitimate, 1 - fraudulent. Furthermore, each instance contains the prediction of each of the 50 experts, following the same convention as the label. We provide every expert’s prediction for every 30K instances in the Bank Account Fraud dataset (https://www.kaggle.com/datasets/sgpjesus/bank-account-fraud-dataset-neurips-2022/versions/1?select=Base.csv) deemed fraudulent by a fraud detection model, thus simulating an “alert-review” scenario, where experts are tasked with reviewing high-risk bank account opening applications.