KAIST researchers develop AI to detect stress via smartphone behavior

Researchers at KAIST have developed an artificial intelligence technology, named ‘SHIRBT-MMF’, capable of identifying daily stress levels by analyzing repetitive smartphone interaction patterns. Rather than monitoring specific app usage or reading private message content, the system focuses on how users interact with their devices, such as the frequency of typing, screen transitions, and patterns in web browsing, messaging, or financial transactions.

To ensure privacy, the technology does not store actual text or call details; instead, it uses hashing and records only behavioral metadata. The research team utilized Multi-Level Sequential Pattern Mining (ML-SPM) and Large Language Models (LLM) to classify activities. In tests involving 26 participants, the AI achieved a stress detection accuracy of 75.9% for short-term data and 73.8% for long-term data, outperforming traditional methods that rely solely on app usage time.

Professor Lee Eui-jin noted that the research views the smartphone as a living space for communication and information seeking rather than just a device for measuring screen time. While the technology is not intended to diagnose clinical mental illnesses like depression, it provides a foundation for stable, long-term mental health monitoring through personalized AI models.