Depression and Suicide Risk Detection From Internet Usage Traces

Loading...
Thumbnail Image

URL

Journal Title

Journal ISSN

Volume Title

Perustieteiden korkeakoulu | Master's thesis

Date

2024-03-11

Department

Major/Subject

Machine Learning, Data Science and Artificial Intelligence

Mcode

SCI3044

Degree programme

Master’s Programme in Computer, Communication and Information Sciences

Language

en

Pages

92 + 13

Series

Abstract

Depression is one of the leading causes of illness and disability worldwide. Over the past few decades, there has been a surge in our reliance on the internet, advancing the prospect of utilizing our online behavior as a diagnostic tool for identifying depression and suicide risk. At the same time, it raises the question of potential associations between internet use and mental health. Previous research on internet usage for mental health assessment pertains mainly to data from mobile devices from small homogeneous populations. This thesis explores the potential of internet usage (IU) features from desktop and mobile devices for depression and suicide risk assessment using a large heterogeneous population of about 900 individuals per device type. This study shows that IU features can distinguish people with no depression symptoms from people with high depression severity with an accuracy of 0.61, which improves to 0.66 when combined with sociodemographic features. The IU features performance for recognizing people with none or minimal depression severity from people with mild or higher depression severity is 0.56, which improves to 0.60 when combined with sociodemographic features. Lastly, the IU features performance for recognizing people presenting suicide risk symptoms is 0.54, which improves to 0.57 when combined with sociodemographic features. In all cases, the sociodemographic features alone achieve the best accuracy, ranging from 0.59 to 0.73. To uncover existing associations between internet usage and depression or suicide risk, this study uses hierarchical mixed-effect models with study participants as random effect to account for individual-level characteristics. The regression analysis reveals that the daily count of application views, the count of application views during the night, the total time spent on chat and messaging platforms, the time spent on message boards and forums, and the number of job-related URLs all have statistically significant positive associations with depression. For suicide risk, it is found that the time spent on chat and messaging platforms, the number of health-related applications, and the number of job-related URLs have positive statistically significant associations with suicide risk severity. Collectively, the results advocate for a comprehensive and inclusive approach to mental health assessment that integrates both traditional sociodemographic factors and emerging internet usage patterns.

Description

Supervisor

Kulshrestha, Juhi

Thesis advisor

Aledavood, Talayeh

Keywords

digital phenotype, internet usage, web-browsing behaviour, depression, suicide risk, app usage

Other note

Citation