A hybrid classification model for identifying college students with depression

• Grant number: 24/12772-2
• Support Opportunities: Scholarships in Brazil - Master
• Start date: April 01, 2025
• End date: February 28, 2026
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Vânia Paula de Almeida Neris
• Grantee: Evandro Yudi Alves Ribeiro
• Host Institution: Centro de Ciências Exatas e de Tecnologia (CCET). Universidade Federal de São Carlos (UFSCAR). São Carlos , SP, Brazil

Abstract

There is a consensus in the field of mental health that, in cases of major depression, a comprehensive model of care and follow-up is necessary, one that does not rely solely on pharmaceuticals. Studies in this area have been seeking support from computational solutions, mainly from the fields of Human-Computer Interaction (HCI) and Artificial Intelligence (AI). Research on Machine Learning (ML) approaches in the field of mental health has explored the use of data from mobile sensors to identify and infer symptoms related to depression. Similarly, works involving Natural Language Processing (NLP) investigate the detection of depressive symptoms from postings on Online Social Networks (OSNs) and other textual forms of self-reporting, such as diaries. Despite the advancements observed in the literature on mental health, there is a noticeable gap in studies involving the combination of sensor time series data with textual data. This strategy of combining different types of data has been explored in ML, multimodal, and hybrid models and has demonstrated superior performance compared to approaches using only one type of data. The hybrid approach is prominent as it is based not only on the combination of different types of data but also on different ML techniques and algorithms. Therefore, this project aims to investigate the construction of a hybrid model for the identification of a Possible Depressive Profile (PDP) using jointly processed mobile sensor data and textual data in a time series. This project will explore an existing computational infrastructure from the Amive - Specialized Virtual Friend project, funded by FAPESP (20/05157-9), which collected data from university students with and without depression. The infrastructure is designed to assist in the identification of PDP using processed sensor data through an ML classifier and OSN data processed by an NLP model. Among the challenges found in the Amive infrastructure, this project is interested in addressing the underfitting problems identified in current solution's classifier and the lack of joint work of sensory and textual data from ML and NLP models. The expected results can contribute to the state-of-the-art in the field by testing fine-tuning strategies for the identification of PDP and proposing a classification solution that considers sensor and textual data together.