An Efficient Advanced Data Integration From Multiple Sources For Fraud Detection Using Etl And Fl-Zsl-Glnrp3n

Abstract

Advanced data integration combines data from multiple sources, which are then transformed and loaded into a unified system for decision-making. However, the existing studies didn’t integrate structured and unstructured data for fraud detection in U.S financial transactions, affecting accurate decision-making. Therefore, this paper presents advanced data integration from multiple sources for fraud detection using ETL and FL-ZSL-GLNRP3N. Initially, the users are registered in the financial app, followed by key generation. While the user logs in to the financial app, the transaction is initiated. Afterward, identity theft mitigation is done. If the verifications are successful, the transaction details are encrypted by utilizing Koblitz Zorro Curve Cryptography (KHCC). Using FL, the fraud detection system is trained in a local model and updated in the global model. In the fraud detection system, ETL is employed for advanced data integration. Here, unstructured review data is structured using a Python library. Afterward, the data are integrated using COMA++. Next, temporal behavior analysis by Discrete Six-Hump Camel Jordan Wavelet Transform (DSHCJWT), word embedding, and feature extraction are performed. Lastly, fraud detection is done by using FL-ZSL-GLNRP3N. Here, Local Interpretable Model-agnostic Qing Explanation (LIMQE)-based DeepXplainer and ZSL are used. If the transaction is fraudulent, then it is stopped; otherwise, the transaction is completed. During testing, reviews obtained from original users and decrypted transaction details are employed for the fraud detection system. The results proved that the proposed model achieved a high accuracy of 98.76%.