Development of Banking Transactions Compromise Detection Model

The final qualifying work is devoted to developing of fraud transactions detection complex approaches based on their attributes using machine learning principles and other analysis methods. As a part of the work, there was conducted a study on bank transaction including analysis of their types, features, and available attributes. Also, the research includes the elicitation of classification and rare events detection methods like logistic regression, naive Bayes, K-nearest neighbors, support vector machine, decision tree, adaptive boosting and random forest. It has resulted in final fraud detection models requirements. Based on conducted analysis several predictive models were developed and then combined in complex analytical approaches of potentially fraudulent transactions detection. The developed algorithms were tested with real data including impersonal information on bank transactions by European cardholders. As a result of testing, these models demonstrated a fairly high predictive ability, as the models were able to detect about 80% of fraudulent operations and reach the level of F-score more than 0.86. The developed complex approaches can be effectively integrated into Fraud Prevention and Fraud Detection systems algorithms of transactions sanity check.