Fraud Detection in Credit Card Transactions Using HDBSCAN, UMAP and SMOTE Methods

Abstract

Credit card abuse and fraud in credit card transactions pose a serious threat to financial companies and consumers. To overcome this problem, accurate and effective fraud detection is essential. In this study, we propose an approach that combines HDBSCAN (Hierarchical Density-Based Spatial Clustering of Applications with Noise), UMAP (Uniform Manifold Approximation and Projection), and SMOTE (Synthetic Minority Over-sampling Technique) methods to detect fraud in credit card transactions. The HDBSCAN method is used to group transactions based on their spatial density, allowing identification of suspicious groups of transactions. UMAP is used to reduce the dimension of transaction data, thus enabling better visualization and more efficient data analysis. In addition, we use SMOTE to overcome class imbalances, namely differences in the number of fraudulent and non-fraudulent transactions. In our experiments, we used. In this experiment, we used a dataset of credit card transactions that included both fraudulent and non-fraudulent transactions. The experimental results show that the proposed approach is able to detect fraud with high accuracy. The HDBSCAN method is able to effectively identify suspicious groups of transactions, while UMAP helps in better understanding and visualization of data. The use of SMOTE has successfully overcome class imbalances, resulting in more balanced fraud detection results between fraud and non-fraud. The results of this study show that the combination of HDBSCAN, UMAP, and SMOTE methods is effective in detecting fraud in credit card transactions. This approach can help financial companies identify suspicious transactions with high accuracy, reduce fraud losses, and improve the security of credit card transactions.

Topics & Concepts