<![CDATA[Polycystic Ovary Syndrome (PCOS) is an endocrine disorder affecting approximately 6–21% of women of reproductive age but is often difficult to detect in its early stages. This study develops a PCOS detection model using machine learning and the Synthetic Minority Oversampling Technique (SMOTE) to address class imbalance between positive and negative classes. Four algorithms were tested: Logistic Regression (LR), K-Nearest Neighbors (KNN), Decision Tree (DT), and Random Forest (RF). The dataset used was obtained from Kaggle and underwent preprocessing, including data cleaning, encoding, and feature selection based on correlation. After balancing the data using SMOTE, standardization was applied to ensure feature scale consistency.Model evaluation was carried out using a confusion matrix as the basis for calculating Accuracy, Precision, Recall, F1-score, Specificity, and Negative Predictive Value (NPV), with Recall as the primary focus. The results show that LR achieved the highest Recall (0.92), while RF demonstrated the best performance balance with Accuracy (0.92), F1-score (0.87), Specificity (0.95), and NPV (0.93). KNN and DT obtained the same Recall value (0.83), although their Precision and F1-score were slightly lower. In addition, a before–after analysis was conducted to evaluate the effect of SMOTE, and the McNemar test was used to assess the statistical significance of performance differences between models. Feature Importance analysis revealed that follicle count, menstrual cycle length, and fast-food consumption patterns are the most influential factors contributing to PCOS risk. These findings indicate that the application of SMOTE significantly enhances model sensitivity and has strong potential to be developed as a decision support system in reproductive health.]]>
