Detecting Human Emotions Using Smartphone Accelerometer Data
Year: 2016
Type: dissertation
Abstract: The work presented in this thesis concerns the classification of emotions based on smartphone accelerometer data. The data is collected from individuals who have been carrying their phone in a pocket while walking. While other methods of emotion detection generally are more intrusive or only give estimations over longer periods of time, using accelerometer data presents an opportunity for real-time and non-intrusive emotion detection. An Android app was developed in order to monitor the smartphone accelerometer of the individuals who participated in the study and occasionally request them to submit their emotional state. This way, data is collected from a natural environment rather than a laboratory setting. The recorded data is then processed and used to train different classifiers to be compared. The machine learning algorithms decision tree, support vector machine and multilayer perceptron are used for this purpose. Emotions are classified in two dimensions: pleasantness and arousal (activation). While the recognition rate for the arousal dimension is promising at 75%, pleasantness is harder to predict, with a recognition rate of 51%. These findings indicate that by only analyzing accelerometer data recorded from a smartphone, it is possible to make predictions of a person's state of activation. Such predictions can be used, especially in conjunction with other methods of emotion detection, to adapt services to the user's emotional state.
Author: Andreas Færøvig Olsen
Language: English
Topic: Emotion and Mood Recognition
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2601889421