Detecting Fraud Transactions in Financial Institutions

Abstract

Detecting fraud and anomalies in financial transactions is crucial in safeguarding institutional assets,
maintaining regulatory compliance and ensuring customers trust in financial system. This study
investigated methods of detecting frauds or anomalies in transactions within financial institutions, a vital
task to prevent financial losses, reduce investigative costs, and comply with regulatory standards. We
compared the efficiency of three statistical models: Logistic Regression, Linear Discriminant analysis
(LDA).and Quadratic Discriminant (QDA), in identifying fraudulent activity. Secondary data of over
280,000 financial transactions from an online website (Kaggle) was used to evaluate each model based on
accuracy, precision, and error rates, for both fraudulent and non-fraudulent classifications. The results
indicated that Logistic Regression outperformed LDA, and QDA, achieving the highest accuracy and
lowest error rate, making it the most effective model among the models considered in the study for fraud
detection in this context.