Final year project · Aug 2025 - May 2026
Stress Detection via Data Analysis of Keystroke Dynamics
An end-to-end pipeline for collecting and analysing typing patterns as a signal for stress.
Idea
Most stress detection tools are either intrusive, reactive, or rely on self-reporting - none of which work well in everyday settings. This project explores whether keystroke dynamics collected during normal computer use can reveal statistically meaningful differences between low and high stress states, without any extra hardware or disruption to the user. We built a browser-based data collection platform, ran participants through controlled typing tasks designed to simulate varying cognitive load, and extracted behavioural features like key hold time, inter-key intervals, and typing speed. Statistical analysis confirmed observable differences between conditions, and exploratory ML classification achieved ~73% accuracy in distinguishing stressed from non-stressed sessions across 60 participants.
