OSIRA, an AI Early Warning System Using Open Source Intelligence for Digital Fraud Prevention

Purpose: This study designs and evaluates OSIRA (Open Source Intelligence Risk Analyzer), an early warning system based on artificial intelligence that integrates Open Source Intelligence (OSINT) and Natural Language Processing (NLP) to identify digital fraud indications before they enter formal reporting mechanisms in a cashless society. Research Methodology: The study uses a design science approach consisting of problem analysis, system design, risk score formulation, and a limited scale proof of concept simulation. The simulation used 540 synthetic texts from 50 template groups, divided into 394 training and 146 testing texts through a group based split, and classified them with Term Frequency–Inverse Document Frequency (TF IDF) features and logistic regression. Results: OSIRA integrates web crawling, text preprocessing, fraud classification, entity extraction, a risk engine, and an early warning module. At a threshold of 0.50, the model achieved 82.88% accuracy, 79.66% precision, 98.95% recall, an F1 score of 88.26%, and a ROC AUC of 0.8528, but its false positive rate reached 47.06%, mainly for normal texts discussing fraud related topics. Conclusions: OSIRA shows initial feasibility as an additional intelligence layer that can shift fraud mitigation from reactive response toward proactive fraud intelligence. Limitations: The evaluation used synthetic rather than real OSINT data, a baseline classifier with limited contextual understanding, and risk score weights and thresholds that have not been empirically calibrated. Contributions: The study offers a conceptual and technical framework for using pre reporting public signals in fraud risk management and provides guidance for banks, digital wallet providers, and regulators.