Polycystic Ovary Syndrome Detection Machine Learning Model Based on Optimized Feature Selection and Explainable...

Area of Science:

• Medical Diagnostics
• Machine Learning Applications
• Women's Health

Background:

• Polycystic Ovary Syndrome (PCOS) is a prevalent global health issue affecting women.
• Early PCOS detection and treatment are vital to mitigate risks of type 2 and gestational diabetes.
• Machine learning (ML) and ensemble learning show potential for improving medical diagnostic accuracy.

Purpose of the Study:

• To develop and evaluate ML models for accurate and trustworthy PCOS diagnosis.
• To enhance model interpretability through local and global explanations.
• To identify optimal feature selection and ML models for PCOS detection.

Main Methods:

• Utilized various ML models including logistic regression, random forest, SVM, and ensemble methods like stacking.
• Employed feature selection techniques and Bayesian optimization for model tuning.
• Addressed class imbalance using SMOTE (Synthetic Minority Oversampling Techniques) and ENN (Edited Nearest Neighbour).

Main Results:

• The Stacking ML model combined with REF feature selection achieved 100% accuracy.
• This performance surpassed other evaluated ML models and feature selection methods.
• Experimental results were validated on a benchmark PCOS dataset with 70:30 and 80:20 data splits.

Conclusions:

• Stacking ML with optimal feature selection offers a highly accurate approach for PCOS diagnosis.
• Model interpretability is key for ensuring trust and effectiveness in clinical applications.
• Advanced ML techniques can significantly aid healthcare systems in managing PCOS and its complications.