Toward mental health prediction using browsing history for predictive and soft labeling
Date
2022
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
Web browsing data is increasingly being explored as a passive window into the daily lives of users to assess levels of internet addiction and other mental health markers. In this work, I determine whether such approaches can be applied to data generated by the HabitLab platform to assess the community of users for signs of anxiety, depression, and loneliness. HabitLab is a Chrome-based browser plugin that offers users tools to monitor and optimize the time they spend on various websites that negatively impact their productivity while passively logging their browsing sessions. As part of an initial explorative study, 66 HabitLab users completed a paid Qualtrics survey during the height of the COVID-19 pandemic in the US. I analyzed their response to several psychometric scales included on the survey and paired their results with their web browsing data. I then developed several features from this data to characterize their behaviors and used Machine Learning techniques (e.g., SVM, Random Forest) to attempt to learn relationships between their responses. I trained both Classification and Regression models, as I had access to both the real valued scores and their interpretations for most of the mental health scales. The results suggest that the models (specifically regression models), are capable of learning some of the scales; achieving over 80% accuracy for predicting Anxiety and Sleep Disturbance. As the long-term goal of the HabitLab team is to transform the plugin into a Digital Wellbeing and Occupational Health platform, these results can be used to inform: (i) future onboarding and demographic intake questionnaires, (ii) efforts to develop features based on web usage data and predictive models of mental health status, and (iii) facilitate anonymous community assessment through soft labeling approaches.
Description
Keywords
Web browsing, Mental health markers, COVID-19 pandemic, Anxiety, Machine learning