A Statistical Classifier for Assessing the Level of Stress from the Analysis of Interaction Patterns in a Touch Screen

Abstract

This paper describes an approach for assessing the level of stress of users of mobile devices with tactile screens by analysing their touch patterns. Two features are extracted from touches: duration and intensity. These features allow to analyse the intensity curve of each touch. We use decision trees (J48) and support vector machines (SMO) to train a stress detection classifier using additional data collected in previous experiments. This data includes the amount of movement, acceleration on the device, cognitive performance, among others. In previous work we have shown the co-relation between these parameters and stress. Both algorithms show around 80% of correctly classified instances. The decision tree can be used to classify, in real time, the touches of the users, serving as an input to the assessment of the stress level.