Enhanced Anomaly Detection in Embedded Payment Systems using Depthwise Separable CNN with Dandelion Optimizer

Due to the prevalence of online payments, it is essential to identify instances of payment fraud. Plans that break rules and regulations within the payment systems not only cost money but also erode customer confidence. Finding these fraudulent transactions can be challenging because of the complexity of fraud and the sheer number of transactions. In this study, a new deep learning technique for identifying online payment fraud is proposed. The classification phase, hybrid feature selection, and sophisticated preprocessing form its foundation. Starting with the IEEE CIS Fraud Dataset and the PaySim Financial Credit Dataset, the suggested approach is implemented. To ensure that all of the data sets have identical scales, a preprocessing step that uses Min-Max scaler normalization is utilized. The optimal qualities are picked using a hybrid approach that combines Gooseneck Barnacle Optimization (GBO) and the Artificial Hummingbird Algorithm (AHA). A depthwise separable convolutional neural network (DS-CNN) is then employed for classification, which facilitates the differentiation between fraudulent and non-fraudulent transactions. Additionally, to increase the detection rate, the Dandelion Optimizer (DO) optimizes the entire system. The aforementioned study indicates that the suggested approach has a high accuracy rate in detecting fraudulent transactions, which means it has almost infinite potential to speed up the entire process. In both datasets, the proposed method achieves an impressive 99% accuracy rate, proving its dependability and efficiency. This exceptional outcome shows how flexible the approach is to complex patterns and data variations, indicating its high level of reliability.