A Comparative Study for Detecting Polycystic Ovary Syndrome Using A Machine Learning Framework

Polycystic Ovary Syndrome (PCOS) is a widespread hormonal disorder affecting numerous women with ovaries, during their reproductive ages, often leading to various complications if undiagnosed or improperly managed. This research aims to improve PCOS detection by utilizing an interpretable machine-learning framework for diagnostic assistance. We utilized a publicly available dataset from Kaggle, featuring 541 samples and 44 attributes. Various preprocessing techniques and feature selection methods, such as Chi-Square, Mutual Information, and Correlation Coefficient, were employed here. Subsequently, we trained multiple machine learning algorithms and subjected them to K-fold cross-validation. Their performance is evaluated and compared by accuracy, precision, recall, and F1 score. Our findings reveal that the CatBoost algorithm, when paired with Chi-Square feature selection, achieved the highest accuracy rate of 93.9%. Our study focuses on investigating important features, comparing algorithm performance, and achieving high accuracy in PCOS detection. The research validates machine learning's effectiveness for non-invasive PCOS diagnosis, laying the groundwork for future medical diagnostics.