A Study on Digital Payment Fraud Prevention System Using Predictive Modelling

Authors:

Mr. Dhananjay Nishane¹, Prof. Prashant Sawarkar²

¹Student, Department of Business Management, SBJITMR, Nagpur

²Assistant Professor, Department of Business Management, SBJITMR, Nagpur

ABSTRACT

The rapid evolution of real-time digital transaction channels has expanded the surface area for highly sophisticated financial fraud. Existing legacy security networks rely heavily on traditional rule-based filters that are operationally rigid and generate high false-positive rates, compromising consumer trust. To mitigate these system inefficiencies, this study proposes an enhanced predictive modelling framework utilizing a robust Random Forest classifier optimized to intercept anomalous financial behaviors within UPI-integrated mobile payment networks. Evaluating the system on a highly imbalanced synthetic transactional dataset, the proposed configuration achieved a precision score of 1.00, completely eliminating false alarms, alongside a recall of 0.48. Crucially, feature analysis reveals that illicit activity is exclusively clustered inside the “TRANSFER” and “CASH_OUT” channels. The findings establish that tree-based predictive modeling can effectively insulate high-velocity settlement networks by isolating malicious capital pipelines before funds are permanently liquidated.

Keywords: Predictive Modelling, Digital Payment Fraud, Unified Payments Interface (UPI), Random Forest Classifier, Machine Learning, Streamlit Dashboard, Data Imbalance.