Hybrid Cognitive Framework for Fraud Detection using NLP and Transaction Behavior Analytics

Abstract

Unified Payments Interface (UPI) presents an intelligent framework for detecting fraudulent activity in high-volume transactions by integrating behavioral analytics with machine learning and Natural Language Processing (NLP) on interaction data. India processes over 12 billion UPI transactions per month with year-on-year growth exceeding 40%, while digital payment fraud results in losses of several thousand crores annually, highlighting the urgency for real-time protection mechanisms. Such systems are essential in application scenarios including peer-to-peer transfers, bill payments, and other instant transactions where approvals occur within seconds and delays can lead to irreversible losses. Traditional manual and rule-based fraud detection approaches rely on static thresholds, delayed human intervention, and limited behavioral understanding, making them unsuitable for dynamic fraud patterns and largescale transaction volumes. The proposed UPI- Shield Fraud Detection (UPI-SFD) system utilizes a structured UPI dataset enriched with text-based behavioral information, which undergoes NLP preprocessing, exploratory data analysis, and numerical feature extraction using Statement Bidirectional Encoder Representations from Transformers (SBERT) to capture semantic intent. Synthetic Minority Over-sampling Technique (SMOTE) is applied to balance class distributions, after which existing classifiers such as Gaussian Naive Bayes, Bernoulli Naive Bayes, and Multinomial Naive Bayes Classifiers are evaluated as baselines, while the Histogram Gradient Boosting classifier is introduced as the proposed model for improved non-linear learning and real-time performance. The system generates two outputs: UPI Shield classes1 identifying transactions as Normal or Fraud, and UPI Shield classes2 classifying transaction types P2P, Bill Payment, or Other, thereby enabling accurate fraud detection and comprehensive behavioral analysis within real-time payment systems.