Classification of psychiatry clinical notes by diagnosis: a deep learning and machine learning approach

Area of Science:

• Artificial Intelligence in Healthcare
• Computational Psychiatry
• Natural Language Processing for Mental Health

Background:

• Accurate classification of clinical notes is vital for mental health diagnostics, particularly for conditions like anxiety and adjustment disorder.
• Evaluating AI model performance requires comparing various architectures and data balancing techniques.

Purpose of the Study:

• To compare traditional machine learning and deep learning models for classifying clinical notes into anxiety and adjustment disorder diagnoses.
• To assess the impact of oversampling strategies (No Oversampling, Random Oversampling, SMOTE) and hyperparameter tuning on model performance.

Main Methods:

• Utilized machine learning models (Random Forest, SVM, KNN, Decision Tree, XGBoost) and deep learning models (DistilBERT, SciBERT).
• Implemented three oversampling strategies: No Oversampling, Random Oversampling, and SMOTE.
• Performed hyperparameter tuning to optimize model accuracy for classifying clinical notes.

Main Results:

• Oversampling techniques generally had minimal impact, except for SMOTE with BERT-based models.
• Hyperparameter tuning significantly improved accuracy across all tested models.
• Decision Tree, XGBoost, DistilBERT, and SciBERT models achieved 96% accuracy.

Conclusions:

• Hyperparameter tuning is crucial for maximizing AI model performance in clinical note classification.
• Both traditional ML and deep learning models demonstrate high potential for AI-assisted mental health diagnostics.
• Findings offer insights into effective AI strategies for mental healthcare applications.