Enhancement in HTM-CLA for credit card fraud detection using H-DS and H-MSC classifier

With digital payments becoming more widespread, credit card usage increased, and the use of cash went down. Alas, this also leads to an increased risk of credit card fraud, which is a significant danger to financial security online. Brain-inspired computing techniques involving intelligent classification methods to enhance fraud detection and false alarm reduction. A Hierarchical Temporal Memory (HTM)-based, biology-inspired approach is used for the detection of fraud in transactional data. Synthetic data that simulates actual transactions was analyzed. The HTM model was implemented, including scalar encoders, spatial poolers, and temporal memory modules. It creates Sparse Distributed Representations (SDRs) to support sequence learning robustly. Two classifiers enhanced the HTM output: H-DS (Hierarchical Decision System) and H-MSC (Hierarchical Multi-Stage Classifier). Both were trained on the entire dataset. The evaluation was based on training performance, measuring accuracy, precision, recall, and fraud detection rate. 99% accuracy has been achieved for the proposed approach.