Hybrid Machine Learning Algorithms for Polycystic Ovary Syndrome Detection

This study focuses on identifying Polycystic Ovary Syndrome (PCOS), which is a serious medical hormonal disorder condition that affects a woman’s ability in childbearing age and causes infertility, diabetes, heart problems, endometrial cancer, and other related health issues as it raises the possibility of long-term problems. Therefore, early detection is of great importance. Taking into account Machine Learning (ML) and ensemble learning algorithms with their superior detection capabilities, especially in the medical field. The Hybrid Random Forest Logistic Regression (HRFLR), combined Extreme Boosting with Random Forest, Linear Support Vector Machine, Light Gradient Boosting Model, and CatBoost model are among the hybrid ML models used in this study. To support the unique techniques, the classification models have been investigated using the PCOS dataset containing physical and clinical parameters of women collected from 10 different hospitals across Kerala, India, and downloaded from the Kaggle repository. The top 14 physical and clinical features were selected to detect PCOS by using the univariate feature selection approach. Follicle No. (R) and Follicle No. (L) were found to be the most significant among all features. The evaluation metrics used to test all the models are Accuracy, Precision, Recall, F1-Score, ROC curve plot, Area Under the Curve Score, and K-fold Cross Validation (CV). Finally, results were discussed and compared to indicate that CatBoost outperforms other models, with an accuracy score of 92% also HRFLR obtained the best CV accuracy with 92% when K = 30, both applied on the top 14 features only using 80:20 data split ratio. Consequently, CatBoost is effective at detecting PCOS patients.