Polycystic Ovary Syndrome Prediction Using Machine Learning: A Comparative Analysis of Classification Algorithms.

Abstract

Background: Polycystic ovary syndrome (PCOS) represents one of the most prevalent endocrine disorders affecting women of reproductive age, with significant implications for metabolic, reproductive, and psychological health. Early and accurate diagnosis remains challenging due to heterogeneous clinical presentations and the complexity of diagnostic criteria.

Objective: This study aimed to develop and compare multiple machine learning algorithms for predicting PCOS diagnosis, evaluating their performance across various metrics to identify the most effective computational approach for clinical decision support.

Methods: A comprehensive dataset containing clinical, biochemical, and anthropometric parameters from patients was analyzed using twelve different machine learning algorithms. The dataset underwent rigorous preprocessing including missing value imputation, feature engineering, and categorical encoding. Models evaluated included Logistic Regression, Support Vector Machine, K-Nearest Neighbors, Naive Bayes, Decision Tree, Random Forest, Gradient Boosting, XGBoost, AdaBoost, Neural Network, LightGBM, and HistGradientBoosting. Performance was assessed using accuracy, F1-score, sensitivity, specificity, and ROC-AUC scores.

Results: Gradient Boosting, XGBoost, and HistGradientBoosting demonstrated superior performance with accuracy of 92.66%, while ensemble methods generally outperformed single classifiers. Gradient Boosting achieved the highest F1-score of 87.87% and ROC-AUC of 95.39%. Random Forest exhibited exceptional specificity at 98.63%, while Naive Bayes showed the highest sensitivity of 94.44%. Traditional machine learning approaches like SVM and Neural Networks showed comparatively limited performance in this context.

Conclusion: Machine learning algorithms, particularly gradient boosting methods, demonstrate substantial potential for accurate PCOS prediction and can serve as valuable tools for clinical decision support, potentially enabling earlier intervention and improved patient outcomes.