Abstract
Introduction
Polyendocrine metabolic ovarian syndrome (PMOS), formerly known as polycystic ovary syndrome (PCOS), is a common endocrine and metabolic condition associated with reproductive, endocrine, and metabolic abnormalities. Predictive modeling provides a data-driven approach for evaluating how multiple clinical factors may contribute to PMOS diagnosis. This study compared logistic regression, k-nearest neighbors (KNN), and support vector machine (SVM) models for predicting PMOS diagnosis using five clinical variables: age, body mass index (BMI), menstrual irregularity, testosterone level, and antral follicle count.
Methods
A publicly available dataset containing 1,000 observations was analyzed in R/RStudio. PMOS diagnosis was used as the binary outcome variable, with 19.9% of observations classified as PMOS-positive. The data were partitioned into 75% training and 25% testing sets. The same training and testing sets were used across all three models, and predictors were centered and scaled for the KNN and SVM models. Classification performance was evaluated using accuracy, sensitivity, and specificity.
Results
All three models demonstrated strong classification performance. Logistic regression achieved 89.56% accuracy, 75.51% sensitivity, and 93.00% specificity. KNN achieved 95.18% accuracy, 89.80% sensitivity, and 96.50% specificity. SVM demonstrated the strongest overall performance, achieving 96.79% accuracy and 99.00% specificity, whereas KNN produced the highest sensitivity. The SVM model correctly classified 241 of 249 observations in the testing set, with eight total misclassifications.
Conclusion
These findings demonstrate that the selected reproductive, endocrine, and metabolic variables were able to discriminate between PMOS-positive and PMOS-negative observations within this dataset. SVM provided the strongest overall classification performance, while KNN demonstrated greater sensitivity for identifying positive cases. Given the class imbalance and use of a single publicly available dataset, these findings should be considered exploratory and require external validation in additional and more diverse populations before clinical application.