Rundensteiner, Elke A.
Paffenroth, Randy Clinton
Using audio and text data from multiple sources, we evaluated the viability of using machine and deep learning to identify depression and anxiety. Machine learning methods using sub-clip boosting achieved an F1 score of 0.81 for depression and 0.83 for anxiety. Our convolutional neural networks and long-term short term memory models achieved F1 scores of 0.55 and 0.68 respectively for depression. As feature engineering, we used topological data analysis to create Betti curves in our machine learning pipeline. Furthermore, we developed a pipeline to generate text messages with deep learning models, for data augmentation purposes.
Worcester Polytechnic Institute
Major Qualifying Project
All authors have granted to WPI a nonexclusive royalty-free license to distribute copies of the work, subject to other agreements. Copyright is held by the author or authors, with all rights reserved, unless otherwise noted.