A hybrid SMOTE and Gaussian mixture model based optimized XGBoost framework for bipolar disorder detection

Area of Science:

• Psychiatric diagnostics
• Machine learning in healthcare
• Computational psychiatry

Background:

• Bipolar disorder (BD) diagnosis is challenging due to symptom variability and data imbalance.
• Delayed diagnosis leads to ineffective treatments and poor patient outcomes.
• There is a need for reliable, data-driven tools for accurate BD identification.

Purpose of the Study:

• To develop a hybrid machine learning framework for improved bipolar disorder identification.
• To enhance diagnostic accuracy and consistency using clinical data.
• To create a scalable and interpretable decision-support system for psychiatric healthcare.

Main Methods:

• A hybrid framework combining class balancing (SMOTE), latent subgroup discovery (GMM), and ensemble learning (XGBoost).
• SMOTE was applied to training data to address class imbalance.
• GMM clustering identified patient subgroups, generating probabilistic features for an optimized XGBoost classifier.

Main Results:

• The proposed model achieved 93% accuracy, 97% sensitivity, 93% precision, and 95% F1-score on an independent test set.
• The framework outperformed baseline classifiers (SVM, Decision Tree, Logistic Regression, Random Forest) by 6-12%.
• The hybrid approach demonstrated high performance, indicating its clinical relevance.

Conclusions:

• Hybrid machine learning, integrating SMOTE, GMM, and XGBoost, offers a robust solution for bipolar disorder identification.
• The developed framework provides a scalable, interpretable, and clinically relevant decision-support system.
• This approach supports early BD screening and personalized treatment planning in psychiatric settings.