AI CALL SCAM GUARD

AI-DRIVEN SCAM DETECTION USING LLAMA – A PREPRINT

Abstract

Nowadays, there is an increased demand for threat detection systems to detect scam calls and protect against fraud due to the continuous rise in online and telephonic scams. Rather than relying solely on human resources for protection against fraud, an AI-based threat detection system can help reliably identify scam or spam calls, thereby reducing the number of fraud incidents and associated losses. In this study, we introduce AI Call Scam Guard, an AI-based system prototype for real-time detection of spam and scam phone calls. The backend prototype integrates a fine-tuned LLaMA model, a large language model, within a Python-based framework to evaluate transcribed call content and classify it as safe or scam. The frontend, proposed as part of the system design, is responsible for audio call recording and transcription, enabling potential end-to-end analysis. This project aims to address the rising concern of phone scams by leveraging natural language understanding (NLU) and machine learning (ML) techniques for proactive threat detection. Preliminary results showcase the promising performance of our prototype compared to conventional user-reporting-based systems.