Detección de síntomas de depresión en Redes Sociales mediante Machine Learning y Análisis Facial: un enfoque integrado

Abstract

Introduction: The early detection of depressive symptoms is a critical challenge in mental health, particularly due to the limitations of traditional diagnostic tools and the growing relevance of digital platforms as sources of behavioral and emotional information.
Objective: To develop an automated system for detecting depressive indicators through the analysis of Instagram content, integrating web scraping techniques, natural language processing, facial emotion recognition, and machine learning algorithms.
Method: A computational approach was employed, combining textual classification and facial emotion analysis. For the textual component, Naive Bayes, Logistic Regression, and Random Forest algorithms were evaluated using a dataset of approximately 200,000 text records from Reddit, which was used exclusively for training and evaluating the text classifier. Facial emotion analysis was performed independently on images extracted from the analyzed Instagram profiles. The system was implemented through a Streamlit interface.
Results: The results showed that the Naive Bayes model achieved the best performance for textual classification, with an accuracy of 90% and high recall in detecting depressive indicators. The integration of text analysis and facial emotion recognition allowed the system to compensate for the limitations of each method when used separately and to generate both individual and global reports.
Conclusions: The proposed system contributes to the development of non-invasive tools for the early detection of depressive symptoms, supporting timely interventions and demonstrating the potential of combining machine learning, natural language processing, and facial emotion recognition in digital mental health contexts.