An Advanced Machine Learning Method Used for Identification of Banking Fraud in Financial Transactions

Abstract

In financial security, identifying credit card fraud is still a major problem because of the wildly disproportionate size of transaction databases and the constantly changing strategies used by criminals. To identify credit card fraud, this study uses a Convolutional Cascaded Neural Network (CCNN) model on a highly imbalanced data set (284,807 anonymized transactions) to get reliable results. The data is subjected to a thorough preparation procedure that addresses missing values, uses PCA to extract features, encodes labels, and uses SMOTE to solve the imbalance of minority classes. The CCNN model is trained to determine if the transactions are genuine or fraudulent once the data is divided into training (70%) and testing (30%) sets. Experiments show that the CCNN significantly outperforms well-known models like Naive Bayes (NB) and K-Nearest Neighbours (KNN), thanks to its remarkable 99.93 percent accuracy, 99.85 percent precision, 99.99 percent recall, and 99.93 percent F1-score. The model demonstrated the potential to be applied as a scalable and reliable solution to the real-world credit card fraud prevention situation, and these results supported the model in offering reliable results in detecting fraudulent behavior with very little false positives and false negatives.