LightSentinel: A Lightweight Anomaly Detection System Leveraging Smart Devices
Yingyuan Yang, Xueli Huang, Sunshin Lee, Jiangnan Li · 2024
In recent years, continuous authentication, a method of ongoing identity verification aimed at enhancing cybersecurity protection, has been receiving increasing attention. It utilizes users’ behavior data sampled from various sensors, but due to the resource limitations of smart devices, both the data and computation (including data processing, transmission, and training) have to be offloaded. However, the offloading will introduce more security vulnerabilities. In this paper, we propose LightSentinel, a lightweight continuous user identification and anomaly detection system. LightSentinel can derive users’ key behavior patterns and detect behavior changes during usage without the need for data and computation offloading, and a probability chain has been established for each user to improve the identification accuracy of the system. In our experiment, we deployed LightSentinel on Android devices and conducted evaluations to assess its identification accuracy, energy consumption, and computational efficiency. The results demonstrate that LightSentinel consumes lower power consumption and requires less computation than other applications and IA schemes. Moreover, its accuracy makes it suitable for deployment as an anomaly detection system on smart devices.