The spread of mobile phones in the world, increasing quantity and quality of sensors embedded in modern smartphones open up new opportunities for more individual and less obtrusive stress analysis in real-life situations. One of the aims of our research is to develop a real-life stress recognition method by measuring behavioral data and context, through gathering data from smartphones, and without the use of additional wearable sensors. The parameters collected from smartphones are audio, accelerometer, gyroscope, external lighting, screen light on/off, and self-reports (current stress level assessment). In a binary classification (stress or relax) we achieved over 81% accuracy using activity level information (accelerometer and gyroscope features) and decision tree algorithms. For a 3-class stress classification (low, medium, high) we achieved a 70% accuracy with the application of all features.
Stress recognition; stress tracker; behavioral data; context; classification; smartphone.