QDImResSpSA: An Enhanced Deep Learning Framework For Credit Card Fraud Detection

Credit Card Fraud Detection (CCFD) is an important task due to the increasing number of digital transactions and the sophistication of fraudulent activity. This paper introduces an improved fraud detection (FD) framework named as DImResSpSA, that combines robust preprocessing, advanced feature extraction, and efficient classification methods. SMOTE is used to handle class imbalance, Min-Max normalisation to scale features evenly, and noise removal to improve data quality. Enhanced Principal Component Analysis (ePCA) is used to identify key fraud-related patterns, and quantum PCA (qPCA) along with quantum kernel-based feature mapping is explored to accelerate high-dimensional pattern extraction and improve representation efficiency. To detect complex fraud signatures, a Densely Connected Improved ResNet with a Spatial Pyramid Pooling (SPP) layer and self-attention mechanism is used, allowing the model to learn multi-scale features and long-range connections efficiently. To classify between fraudulent and legitimate transactions, the learnt representations are input into a fully connected layer, which is followed by a SoftMax classifier. The RMSProp optimisation algorithm is used to provide efficient convergence and steady training. The proposed framework enhances fraud pattern recognition, improves classification performance, and offers a viable solution for CCFD.