Machine Learning Approaches for Mental Health Assessment and Intervention in Post-Pandemic Populations

Abstract

The COVID-19 pandemic increased anxiety, depression, and post-traumatic stress disorder around the world. It also revealed gaps in services and led to a rapid rise in the use of machine learning (ML) tools for assessment and support. This scoping review examined post-2020 evidence on ML-based screening, prediction, and intervention in populations affected by the pandemic. Following the Population-Concept-Context framework and PRISMA-ScR, searches in APA PsycINFO and MEDLINE/PubMed found 476 records. After removing duplicates and screening, 19 studies were included. Ten studies validated machine learning models using data from social media, surveys, smartphone or wearable sensors, speech, or electrocardiograms. Nine studies looked at machine learning-enabled chatbots in community or clinical settings. Screening and prediction models showed good results for anxiety, depression, and post-traumatic stress disorder, with multimodal approaches often achieving the best outcomes. Chatbot interventions were practical and well-received, leading to small reductions in depression, anxiety, or loneliness. However, these effects often matched those of self-help controls and depended on ongoing engagement. From all methodologies, the primary challenges have included privacy concerns, potential cultural and linguistic biases, insufficient external validation, and evolving datasets. In general, machine learning approaches offer valuable alternatives for the detection and provision of low-threshold support after the pandemic. Nevertheless, safe scaling requires validation on equity, privacy-preserving designs, transparent reporting, and appropriate practical implementation within routine care in order to establish trust, governance, and clinical acceptance.