A Machine Learning Approach for Efficient Identification and Severity Grading of PCOS and PCOD Using Optimized Features

Infertility, metabolic issues, and hormone imbalance are common symptoms of PCOS, a common endocrine illness affecting women of reproductive age. A various machine learning technique are used in the research for PCOS severity grading and prediction. Recursive Feature Elimination (RFE) was used to choose features first, and then Random Forest and Logistic Regression, two supervised classifiers, applied. The models' efficacy in predicting PCOS was validated by their strong accuracy and AUC ratings. Anti-Müllerian Hormone (AMH) was used as a crucial grading marker after patients were categorized into severity categories (Severe, Moderate, and Low) based on clinical criteria using unsupervised clustering methods, specifically K-Means and Agglomerative Clustering. Well-separated clusters were shown by the silhouette scores used to evaluate the clustering models. A comprehensive framework for early PCOS detection and phenotypic grading is provided by the combination of supervised and unsupervised techniques, which also offers insightful information for individualized treatment plans.