Polycystic Ovary Syndrome (PCOS) Detection via Voting Ensemble Machine Learning Technique: A Multi-Model Approach with Optimized Feature Selection

Polycystic Ovary Syndrome (PCOS) is a common endocrine disorder that affects individuals with ovaries, characterized by irregular menstrual cycles, elevated levels of androgens (male hormones), and polycystic ovaries. In individuals with PCOS, the ovaries often produce excessive amounts of androgens, leading to disruptions in follicular development and ovulation. This hormonal imbalance prevents the normal maturation and release of eggs, resulting in the formation of cysts in the ovaries. Managing PCOS typically involves lifestyle modifications, such as weight management, regular exercise, and a balanced diet, along-side pharmacological interventions like oral contraceptives, anti-androgen medications, and insulin-sensitizing drugs to regulate hormonal imbalances and alleviate symptoms. Machine Learning (ML) has become a valuable tool in achieving this goal. This study proposes a machine learning based framework for PCOS prediction, employing 23 key features. Nine classifiers—Logistic Regression, Stochastic Gradient Descent, Decision Tree, AdaBoost, Gaussian Naive Bayes, Multilayer Perceptron (MLP), K-Nearest Neighbors, Gradient Boosting, and extreme Gradient Boosting (XGBoost) are leveraged to develop an accurate predictive model for PCOS detection. To enhance predictive performance, an ensemble meta-classifier is implemented, aggregating outputs from heterogeneous ML classifiers via weighted voting. This study introduces a novel weighted voting ensemble classifier combined with ANOVA-based feature selection, achieving state-of-the-art accuracy (94.34%) for PCOS detection and identifying 23 key biomarkers, significantly outperforming existing methods through optimized model integration and robust feature engineering.