Enhancing Transaction Fraud Detection with a Hybrid Machine Learning Model

Transaction fraud detection is a critical issue in the financial industry, with fraudsters constantly developing more sophisticated methods to bypass security measures. Traditional fraud detection mechanisms often fall short in adapting to the evolving tactics of fraudsters, creating vulnerabilities in financial transactions. This paper introduces an innovative method to enhance the precision of fraud detection by incorporating cutting-edge machine learning algorithms. Utilizing detailed transaction data from Vesta Corporation, we develop a hybrid model that merges the strengths of LightGBM and Keras neural networks with Focal Loss to tackle the specific challenges presented by skewed data distributions commonly found in fraud detection. This model is designed to improve the detection capabilities of systems, particularly in identifying fraudulent transactions, which are typically rare but have significant impacts. Our empirical findings show marked improvements in detecting fraudulent activities, highlighting the efficacy of our proposed model in strengthening financial security frameworks. This advancement holds great promise for protecting both consumers and businesses from the adverse effects of transaction fraud.