A Robust UPI Fraud Identification Scheme over Digital Money Transactions using Learning Powered Classification Principles

The surge in the number of digital transactions has also led to a huge point in fraudulent activities, particularly in the Unified Payments Interface (UPI), a real time mobile payment system. To identify UPI QR Scan code and UPI-ID fraud, we present a modified deep belief network (M-DBN) prediction model. The system uses a dual phase verification approach involving the use of reliable third party mobile number fraud detectors coupled with the use of a cloud based data privacy validation. The model first collects the sender's mobile number and message content, then sends the collected message content via a data privacy oriented cloud platform to a third party verification service. The second stage is predicting scam score with the current checkpoint coefficients and assign scam score threshold dynamically using real time criteria; years of experience of the QR Scanner badge and verification of third party scam confirmations by double checking. We used perceptrons to identify specific objects and backpropagate error values in order to minimize error values, using the M-DBN model. By stacking Restricted Boltzmann Machines (Res-BMs), and fine tuning the resultant deep network by gradient descent and back propagation, the final model can learn some very appealing hierarchical features and improve its prediction property, even though it may have a complicated initialization. According to the model's confusion matrix, the M-DBN model achieved accuracy as high as 98.4%, precision of 98.1%, recall at 97.8%, and an F1-score of 98.2%. The results show that the proposed method is effective for detecting fraudulent UPI transaction.