Abstract
In recent years, numerous studies have explored the use of machine learning algorithms for supporting applications in social and clinical psychology. In particular, there is an increasing prevalence of smartphone-based techniques for collecting data through embedded sensors and efficient in-situ questionnaires. Models are then built to explore the patterns between these data types. In this paper, we study the application of machine learning for the task of predicting mental states of adverse valence, based on the Photographic Affect Meter data. We present a technique for daily aggregation, which is designed to detect significant negative events. A variety of features is used as input, including GPS-based metrics and features assessing social interactions, sleep and phone usage. Experimental evidence is presented, which suggests that machine learning algorithms could successfully be employed for such a prediction task. © 2019 Association for Computing Machinery.