An efficient framework based on optimized CNN-RNN for online transaction fraud detection in financial transactions

Financial fraud is becoming a serious worry in a world where wireless communications are essential, aimed at transmitting enormous amounts of information while guarding against interference. This paper presents an efficient conceptual architecture (ECA) to detect and flag fraudulent transactions with consideration of various financial transactions. An accurate and robust model has been developed for detecting fraud in online payments. In addition to addressing the dynamic nature of fraudulent behavior and the intolerability requirements that are essential for financial institutions, the suggested system model aims to improve the accuracy of online fraud detection. Initially, the online data is collected and considered as the feature set. After that, the pre-processing step is considered, which consists of normalization and filling in missing parameters. The collected database may contain the label imbalance issue. To solve this issue, the Synthetic Minority Over-Sampling Technique (SIvIOTt) is utilized. To extract the features and identify fraudulent payments, the Optimized Convolutional Neural Network-Recurrent Neural Network (OCNN-RNN) is developed. In this architecture, the CNN layer is utilized to select the features from the input data, and BiLSTM is utilized for sequence detection to obtain the required outcomes. Additionally, the DNN is utilized to optimize the error and loss by using the Enhanced Gazelle Optimization Algorithm (EGOA). Finally, the proposed architecture is used for online fraud detection. The proposed method is implemented in Python, and performances are evaluated by performance measures.