Predicting TCM patterns in PCOS patients: An exploration of feature selection methods and multi-label machine...

Area of Science:

• Biomedical Informatics
• Computational Medicine
• Traditional Chinese Medicine Research

Background:

• Traditional Chinese Medicine (TCM) offers individualized Polycystic Ovary Syndrome (PCOS) treatment via pattern differentiation.
• Subjectivity in TCM diagnoses can lead to inconsistent outcomes.
• Machine learning (ML) integration can provide objective support for TCM diagnoses.

Purpose of the Study:

• To evaluate feature selection techniques and ML algorithms for classifying TCM patterns in PCOS.
• To develop an effective predictive model for enhancing diagnostic standardization.
• To support personalized treatment strategies for PCOS patients.

Main Methods:

• Utilized a dataset of 432 PCOS patients with five distinct TCM patterns.
• Compared Variance Thresholding (VT) with advanced feature selection methods including Recursive Feature Elimination with Cross-Validation (RFECV).
• Evaluated ML algorithms: Support Vector Machine, Logistic Regression, Extreme Gradient Boosting (XGBoost), and Artificial Neural Networks.

Main Results:

• Variance Thresholding reduced features from 224 to 174.
• RFECV proved most effective, identifying 67 key features.
• XGBoost achieved the highest performance with RFECV features: 0.7870 accuracy, 0.9519 F1 score, and 0.0481 Hamming loss.

Conclusions:

• The RFECV-XGBoost model effectively classifies TCM patterns in PCOS patients.
• Precise feature selection is crucial for ML model performance in TCM diagnostics.
• ML significantly advances TCM pattern diagnostics, promoting precise and personalized healthcare.