Comprehensive Study of Machine Learning Approach for Fraudulent Identification in Real-Time Financial Banking Systems

Abstract

Financial fraud, which is commonly defined as applying fraudulent techniques to secure funds, has, in recent years, become a significant issue for businesses and organizations. Efforts to weed out such scams by existing means like human inspection and checks are tedious, expensive and subject to inaccuracies. In the near future, advances in artificial intelligence may enable more sophisticated machine learning algorithms to search through massive amounts of financial data for signs of fraud. This study suggests a thorough method for identifying financial fraud using the highly unequal class-marked IEEE CIS Fraud Detection dataset. A Light Gradient Boosting Machine (LightGBM) and a Convolutional Neural Network (CNN) are the two models that are subsequently trained and assessed on a balanced dataset. The results indicate that CNN model is reliable and significantly higher than LightGBM model, CNN model is 99.73% and its accuracy is higher at 99.91%. Moreover, the CNN model has low false negatives and zero false positives in the confusion matrix, which shows its usefulness in the classification of fraud transactions. The results confirm the effectiveness of CNNs to perform this operation, which gives a solid and significantly effective solution to the problem of real-time financial fraud detection.