Depression detection with machine learning of structural and non-structural dual languages.

Area of Science:

• Computational linguistics
• Mental health informatics
• Social media analysis

Background:

• Depression significantly impacts mental well-being and daily functioning.
• Social media use is increasing globally, with users expressing themselves in regional languages like Roman Urdu in Pakistan and India.
• Previous research has not effectively utilized Roman Urdu for depression prediction, despite its prevalence.

Purpose of the Study:

• To create a novel Roman Urdu dataset for depression risk prediction.
• To evaluate the efficacy of machine learning models in identifying depression using dual-language data (Roman Urdu and English).
• To enhance depression detection in Asian populations through language-specific analysis.

Main Methods:

• Merged datasets comprising manually converted Roman Urdu comments from Facebook and English comments from Kaggle.
• Employed machine learning models: Support Vector Machine (SVM), Support Vector Machine Radial Basis Function (SVM-RBF), Random Forest (RF), and Bidirectional Encoder Representations from Transformers (BERT).
• Classified depression risk into three categories: not depressed, moderate, and severe.

Main Results:

• The Support Vector Machine (SVM) model demonstrated the highest accuracy at 0.84% in predicting depression risk.
• SVM outperformed other tested models, including SVM-RBF, RF, and BERT.
• The study highlights the significant contribution of Roman Urdu in depression prediction when combined with English.

Conclusions:

• The developed Roman Urdu dataset and machine learning approach show promise for improving depression risk assessment in Asian countries.
• SVM is an effective model for classifying depression severity using mixed-language social media data.
• This research contributes to a more nuanced understanding of mental health in digital contexts, particularly for non-English speaking populations.