Unsupervised Classifications of Depression Levels Based on Machine Learning Algorithms Perform Well as Compared to...

Area of Science:

• Psychiatry and Mental Health
• Computational Psychology
• Educational Psychology

Background:

• Traditional depression screening relies on static population norms, struggling with temporal and regional variations.
• Adjusting norm-based cutoffs is infrequent and insufficient for dynamic population changes.

Purpose of the Study:

• To develop an unsupervised machine learning approach for depression classification.
• To overcome the limitations of traditional norm-based depression screening methods.

Main Methods:

• Utilized k-means clustering on data from 8,063 Chinese middle and high school students.
• Generated four levels of depressive symptoms.
• Evaluated classification validity against norm-based methods using robustness, performance metrics (accuracy, AUC, sensitivity), and convergent construct validity.

Main Results:

• The unsupervised machine learning approach successfully generated four levels of depressive symptoms.
• The developed automatic classification system demonstrated comparable performance to traditional norm-based methods.
• The system showed robustness and good convergent construct validity.

Conclusions:

• Unsupervised machine learning offers a promising alternative for depression classification, addressing limitations of static norm-based approaches.
• This method provides a more adaptable and potentially more accurate tool for large-scale depression screening in diverse populations.