AI Detects Stress by Analyzing Routine Smartphone Behaviors, KAIST Researchers Achieve 75.9% Accuracy

KAIST Researchers Develop AI System That Detects Daily Stress

KAIST researchers have developed SHIRBT-MMF, an AI system that detects daily stress by analyzing routine smartphone behaviors rather than tracking which apps people use. Led by Professor Uichin Lee from the Department of AI Computing, the team focused on identifying repeated smartphone behaviors—what users do instead of where they do it—to create a more stable approach to digital mental health monitoring.1

1

The technology addresses a critical limitation in previous smartphone-based stress research. While earlier methods analyzed screen time, app launches, and typing behavior, these approaches became unreliable when apps changed or users switched platforms. The KAIST researchers recognized that even if someone switches from KakaoTalk to LINE or WhatsApp, the underlying activity—messenger chatting—remains constant.2

2

How Smartphone Human Interaction Routine Behavior Tasks Work

The research team defined these recurring patterns as Smartphone Human Interaction Routine Behavior Tasks (SHIRBTs). Using their independently developed Multi-Level Sequential Pattern Mining (ML-SPM) technique, they identified recurring screen-transition patterns across multiple days. A large language model (LLM) then assigned easily understandable labels to these behaviors, such as "messenger chatting," "web browsing," "financial payment," and "reading/writing email."1

1

The system achieved more than 95% accuracy when evaluating whether the same behavior could be consistently classified across different email, web browser, and messenger apps. This behavior-focused approach proved more resilient to changes in the app environment compared to conventional methods.2

2

Privacy Protection Through Data Hashing

To protect privacy, SHIRBT-MMF does not store actual text or message content. The system records only behavioral information—whether text was entered, how often, and screen transitions. Information about message recipients or call partners undergoes data hashing, converting original information into values that do not directly reveal it. This design allows the AI to identify daily stress levels without compromising user privacy.1

1

Analyzing Routine Smartphone Behaviors to Identify Daily Stress Levels

The research team analyzed real-life data from 26 university students and office workers in their 20s and 30s. Among participants with both self-reported stress assessments and smartphone records, 16 short-term participants provided an average of 30.5 days of data, while 10 long-term participants contributed an average of 105.1 days. The researchers built personalized models that distinguished between days with and without stress based on participants' daily self-assessments.2

2

A total of 29 types of repeated smartphone behaviors were identified, with each person routinely repeating about seven behaviors on average. The research team used personalized artificial intelligence models trained on these behavioral patterns to determine whether someone was experiencing stress on a given day.1

1

Source: News-Medical

Related Stories

AI therapist reads smartwatch signals to detect emotional distress before you ask for help

AI-Powered Smartphone Apps Show Promise in Detecting Depression Through Facial and Eye Cues

Researchers Develop AI-Powered 'Digital Thermometer' to Track Social Media-Induced Stress in Real Time

AI Chatbots Display 'Anxiety' in Response to Traumatic Prompts, Study Finds

Stress Detection Accuracy Outperforms Conventional Methods

The best-performing model achieved an average stress detection accuracy of 75.9% for short-term participants and 73.8% for long-term participants—correct roughly three out of four times. These results were 5.5 percentage points and 3.5 percentage points higher, respectively, than the conventional approach of analyzing app usage using the same AI technique.2

2

Performance remained strong even when the number of behaviors analyzed was reduced. Using only two behaviors—messenger chatting and financial payment—or even analyzing messenger chatting alone produced higher average classification accuracy than conventional methods. In long-term data, analytical indicators based on repeated smartphone behaviors changed less over time and maintained a more stable structure than conventional approaches.1

1

Implications for Digital Mental Health Monitoring

The study, published in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies, suggests significant potential for developing digital mental health monitoring technology. Once an individual's repeated behaviors are identified, focusing data collection and analysis on those relevant behaviors could reduce the burden of data collection while maintaining effectiveness.2

2

Source: Tech Xplore

Professor Uichin Lee explained the team's perspective: "We viewed the smartphone not simply as a device for measuring usage time, but as a space in everyday life where an individual's communication, information seeking, and financial activities are repeatedly carried out." This approach makes it easier to understand which behaviors drive stress classification and remains relatively stable even when the app environment changes.1

1

The technology could enable continuous, unobtrusive monitoring of mental health states through devices people already use daily. As smartphone usage patterns evolve and new apps emerge, the behavior-focused methodology positions SHIRBT-MMF to maintain effectiveness where app-specific approaches might fail. Watch for potential commercial applications in workplace wellness programs and personal health tracking platforms.