AUTHOR=Taskynbayeva Meruyert , Gutoreva Alina 

TITLE=Machine learning approaches to anxiety detection: trends, model evaluation, and future directions

JOURNAL=Frontiers in Artificial Intelligence

VOLUME=Volume 8 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/artificial-intelligence/articles/10.3389/frai.2025.1630047

DOI=10.3389/frai.2025.1630047

ISSN=2624-8212

ABSTRACT=BackgroundAnxiety is a pervasive mental health disorder with severe implications for individual wellbeing and societal productivity. The contemporary rise of anxiety, particularly among youth in digitally-saturated environments, underscores a critical need for advanced predictive tools to facilitate early intervention and mitigation. While machine learning (ML) holds significant promise in this domain, a comprehensive synthesis of its application in anxiety prediction, along with a critical evaluation of methodological trends and gaps, is only emerging in the literature. The main idea of the current systematic review is to bridge the understanding of current ML applications in mental health with the critical needs for enhanced diagnostic precision, personalized interventions and prevention.ObjectivesThis systematic review aims to systematically synthesize research on ML approaches to predicting anxiety, critically evaluating the algorithms, features, and validation techniques employed across studies. The objective is to identify prevailing ML techniques, assess their performance, and highlight crucial methodological trends, existing gaps, and their implications for effective early intervention and real-world deployment.Eligibility criteriaStudies included had to apply machine learning techniques to predict anxiety or its severity using either clinical or behavioral datasets. Exclusion criteria included non-English language papers, reviews, older or previously reviewed publications, and those not specifically targeting anxiety. We focus on questionnaire research, but also discuss multimodal fusion techniques.Information sourcesWe searched the Scopus database and Google Scholar for articles published between 2018 and 2025 using combinations of keywords including “anxiety prediction,” “machine learning,” and “mental health.” The last search was conducted in July 2025.Risk of biasStudies were screened in two phases: (1) by verifying the presence of relevant keywords in the main body, and (2) by reviewing title, introduction, and conclusion to ensure alignment with anxiety prediction via ML. Studies relying solely on self-reported metrics or with unclear algorithmic transparency were noted for potential bias.ResultsA total of 19 studies were included, encompassing 44, 608 participants. GAD-7 and DASS-21 were the most commonly used diagnostic instruments. ML techniques such as Random Forest and Gradient Boosting achieved the highest predictive accuracy, with some studies reporting up to 98% accuracy. Metrics like F1-score, AUC, and specificity were commonly reported.Limitations of evidenceExisting studies display a range of methodological and conceptual limitations that constrain their generalizability and clinical utility. The review identified significant methodological limitations hindering generalizability and clinical utility, including reliance on small, homogeneous samples, which raises concerns about overfitting and population bias. Furthermore, common issues include a lack of external validation, inconsistent evaluation metrics, and the “black-box” nature of many ML algorithms, which impedes clinical trust and adoption.InterpretationThe findings support the effectiveness of machine learning for anxiety detection and prediction, particularly in early intervention contexts. The integration of explainable ML and diverse, clinically validated data is necessary for real-world deployment. The existing body of research also shows a notable scarcity in studies predicting anxiety before symptom manifestation. These insights emphasize the critical need for integrating explainable ML (XAI) and utilizing diverse, clinically validated datasets to enable real-world deployment and proactive mental health support.