Language-Based Detection of Depression with Machine Learning: Systematic Review and Meta- Analysis

Area of Science:

• Computational linguistics
• Psychiatry
• Artificial Intelligence

Background:

• Early depression detection is crucial for effective intervention.
• Natural language processing (NLP) and machine learning (ML) offer automated methods for depression detection from text.
• Evidence on the diagnostic performance of these automated methods is limited.

Purpose of the Study:

• To systematically review and meta-analyze studies on NLP and ML for depression detection from spoken or written language.
• To evaluate the overall diagnostic performance and identify factors influencing accuracy.

Main Methods:

• Systematic review and meta-analysis of studies.
• Searched six electronic databases and additional sources, including 123 full-text articles.
• Quantitative synthesis of 50 independent studies, analyzing pooled accuracy, precision, recall, AUC, and balanced accuracy.

Main Results:

• Pooled accuracy across 43 studies was 0.80 (95% CI, 0.76-0.83).
• Subgroup analyses revealed significant differences in performance based on language, text source, feature type, and classifier (p < .001).
• Accuracy was highest with structured clinical interviews, non-English languages, and linguistic or embedding-based features.

Conclusions:

• Automated depression detection from text shows promising performance but with substantial heterogeneity.
• Performance is influenced by language, data source, feature extraction, and model type.
• Methodological standardization and validation are needed for clinical application.