A Machine Learning-Based Predictive Framework for Detecting and Mitigating Online Payment Failures in E-Governance Portals

Abstract

The Digital India initiative has led to rapid growth of digital payments throughout India during the last few years. The Unified Payments Interface UPI serves as the primary factor that drives this expansion. It has become a widely used platform for real-time transactions. UPI processed more than 18.3 billion transactions during March 2025. E-governance portals still experience transaction failures despite their rapid growth. In some cases, money is deducted from a user’s bank account but service is not delivered successfully. This creates problems for citizens because it decreases their trust in online systems. The study presents a machine learning-based framework as the solution for this problem. This framework aims to predict potential transaction failures which it will reduce through real-time monitoring. System uses ensemble learning algorithms which include XGBoost and Random Forest. These mention algorithms help analyze different factors that may affect payment success. The analysis process considers multiple features. It includes network latency, system load patterns, and past merchant performance. Created system uses these factors to determine the transaction risk level. The framework proposed follows a hybrid architecture. It includes an offline model that performs risk scoring. The system includes an online engine that handles mitigation processes. The system uses intelligent payment routing with adaptive retry mechanisms to manage risks. The experimental results show that positive improvements have taken place. The framework can increase transaction approval rates by about 8.7%. The system will decrease the stress that students and users experience when their money gets stuck because of online transaction failures.