Enhanced Diagnosis of Polycystic Ovarian Syndrome Using Stacking Classifier with Random Forest As a Meta Learner

Polycystic Ovary Syndrome (PCOS) is considered a very serious health problem among the women of childbirth age. Early and accurate ‎detection is essential for effective management but remains challenging due to symptom variability and subjective traditional diagnostics. ‎Machine learning (ML) algorithms have recently emerged as promising tools for PCOS diagnosis by analyzing multidimensional clinical, ‎hormonal, and imaging data. This study evaluates and compares multiple supervised ML methods—including Gradient Boosting (XGBoost, ‎LightGBM, and CatBoost), AdaBoost, Logistic Regression, Naive Bayes, Random Forest, Decision Tree, Support Vector Machine (SVM), ‎and Stacking classifiers—on a publicly available dataset of 541 patients. Various evaluation measures like Receiver Operating Characteristic ‎curve (AUC-ROC), recall, accuracy, F1-score, and precision are used for the evaluation. Stacking classifier outperforms all single models ‎used in this research. The Stacking classifier with Random Forest as meta-learner achieved 98% accuracy with AUC of 98%. These find‎ings demonstrate that advanced ensemble ML models can robustly detect PCOS, with potential for integration into clinical workflows for ‎early, objective diagnosis. Future work should focus on multi-center validation and explainable AI to enhance clinical trust and deployment‎.