2023 Summer Undergraduate Research Program (SURP) Symposium
Location
ScholarSpace, Rod Library, University of Northern Iowa
Presentation Type
Poster Presentation (UNI Access Only)
Document Type
poster
Keywords
Social networks--Computer programs; SocioApp--Design and construction;
Abstract
SocioApp is a smartphone application designed to improve student mental health by promoting social interactions using audio data analysis. Initially, the app used a machine learning model to identify users based on their audio features, but it was overfitting and only achieved a validation accuracy of 35.7%. To improve performance, the model's loss function was changed from categorical cross entropy to binary cross entropy and regularization techniques were introduced. These adjustments, which moderated the speed of learning from the training data, significantly increased the validation accuracy to around 70%.
Start Date
28-7-2023 11:00 AM
End Date
28-7-2023 1:30 PM
Event Host
Summer Undergraduate Research Program, University of Northern Iowa
Faculty Advisor
Dheryta Jaisinghani
Department
Department of Computer Science
Copyright
©2023 John Brustkern, Dr. Dheryta Jaisinghani, and Dr. Sarah Diesburg
File Format
application/pdf
Recommended Citation
Brustkern, John; Jaisinghani, Dheryta; and Diesburg, Sarah, "SocioApp: Neural Networks to Detect How Social Are You?" (2023). Summer Undergraduate Research Program (SURP) Symposium. 20.
https://scholarworks.uni.edu/surp/2023/all/20
SocioApp: Neural Networks to Detect How Social Are You?
ScholarSpace, Rod Library, University of Northern Iowa
SocioApp is a smartphone application designed to improve student mental health by promoting social interactions using audio data analysis. Initially, the app used a machine learning model to identify users based on their audio features, but it was overfitting and only achieved a validation accuracy of 35.7%. To improve performance, the model's loss function was changed from categorical cross entropy to binary cross entropy and regularization techniques were introduced. These adjustments, which moderated the speed of learning from the training data, significantly increased the validation accuracy to around 70%.
