Fraud detection by machine learning algorithms

Known as a threat to organizations, fraud, not only destroys their reputation by creating an untrustable environment, but also causes bankruptcy. Fraud as an enterprise risk to be managed can have many social and economic effects in the case of not handled. It is inevitable for organizations in a range of sectors, especially in the banking and finance sector, to invest in technology based on data analytics besides qualified human resources to prevent and detect frauds before reaching substantial damages, and to ensure organizational sustainability. The study aims to develop a predictive model through a machine learning algorithm based on demographic, professional, and financial features for detecting frauds in the banking and finance industry, and to extend the literature in this field by proposing a machine learning-based model. As a research methodology, whether employees committed fraud was tested by classification algorithms, Logistic Regression, Random Forest, Decision Tree, Naive Bayes, and Support Vector Machine with the features of the bank employees worked between 2017 and 2021, and the performance metrics of these algorithms were compared. It is concluded that the Logistic Regression model using the features of Age, Gender, Education Level, Working Time in the Unit, Annual Leave Rate, Personal Credit Rating, Number of Loan Applications, and Credit Card Utilization Rate has the highest accuracy rate and F-Score value.