Prediction of Phone Fraud Risk Using Machine Learning through in-Home Experiments.

Abstract

Phone fraud is increasing dramatically each year in Japan, resulting in the loss of 72.2 billion yen in 2024. Although language processing-based prevention techniques are good way to reduce fraud, fraudsters are adaptive and usually find ways to circumvent such measures. Therefore, we focused on the psychological state of the victim and developed a general-purpose AI for estimating fraud risk using a non-contact sensor for privacy. However, the previous experiments were conducted in the laboratory, it was unclear whether it can be used at home. In this study, we measured psychological and physiological changes in 22 elderly people who experienced conversations simulating phone fraud in their homes to develop a model used in the home. Using this and previous experiment results, we developed a fraud risk estimation model with high accuracy. This technology contributes to the prevention of phone fraud by devices that can be installed in homes.