Financial Fraud Detection Using Value At Risk With Machine Learning In Skewed Data

Abstract

Financial fraud has become a major challenge for banking and financial institutions due to the rapid growth of digital transactions and online payment systems. Detecting fraudulent activities is particularly difficult when transaction datasets are highly imbalanced or skewed, where legitimate transactions significantly outnumber fraudulent ones. Traditional rule-based systems often fail to identify complex and evolving fraud patterns, making it necessary to adopt intelligent data-driven approaches. This study presents a machine learning-based financial fraud detection framework that incorporates the concept of Value at Risk (VaR) to improve the identification of suspicious transactions in skewed datasets. The proposed approach utilizes VaR as a financial risk indicator to estimate potential losses in transaction amounts and integrate it as an additional feature in the fraud detection process. By combining risk-based financial analysis with machine learning techniques, the system is able to capture abnormal transaction behaviors more effectively. The dataset undergoes preprocessing steps including data cleaning, feature selection, categorical encoding, and normalization to prepare the data for model training. Multiple machine learning algorithms such as Logistic Regression, Gradient Boosting, and XGBoost are employed to classify transactions as fraudulent or legitimate. Since fraud detection datasets are typically imbalanced, appropriate techniques such as class weighting and stratified data splitting are used to address skewed data distribution and improve model performance. The models are evaluated based on accuracy and predictive capability, and the best-performing model is selected for final fraud prediction. Experimental results demonstrate that incorporating Value at Risk as a feature enhances the detection capability of machine learning models in identifying risky transactions. The proposed framework provides an effective and scalable solution for financial fraud detection by integrating financial risk metrics with machine learning techniques. This approach can help financial institutions detect fraudulent transactions more accurately, reduce financial losses, and strengthen the security of digital financial systems