Abstract

The growing number and complexity of fraudulent transactions make detecting credit card theft a crucial task in the banking industry. The identification of fraud has made extensive use of traditional machine learning techniques including K-Nearest Neighbor (KNN), Naïve Bayes, SVM, and Logistic Regression. However, these models frequently encounter difficulties when dealing with the temporal correlations and sequential patterns present in transaction data. Long Short-Term Memory (LSTM) networks and Support Vector Machine (SVM) classifiers are combined in this study's hybrid technique to enhance fraud detection capabilities. The LSTM network generates high-level feature representations by efficiently capturing temporal correlations and sequential patterns in transaction sequences. An SVM, which offers strong decision boundaries and improved generalization on unbalanced datasets, is then used to classify these features. The suggested LSTM ? SVM hybrid model performs better than KNN, Naïve Bayes, regular SVM, and Logistic Regression in terms of accuracy, precision, recall, and F1-score, according to experiments done on a benchmark credit card fraud dataset. The findings show that using temporal sequence modeling in conjunction with SVM classification greatly improves the identification of fraudulent activity, which makes it a viable strategy for practical financial security applications.