Feature Selection Strategies for Enhancing the Accuracy for Detecting Polycystic Ovary Syndrome (PCOS) Health Problem

Abstract

A hormonal condition called Polycystic Ovarian Syndrome (PCOS) results in larger ovaries with tiny cysts on the margins. Although the exact etiology of Polycystic Ovary Syndrome is unknown, it may be a result of both hereditary and environmental factors. One of the endocrine diseases that most frequently affect women of reproductive age is Polycystic Ovary Syndrome (PCOS). Artificial intelligence (AI)-based machine learning models has the capacity to classify and predict the potential for PCOS condition. The dataset used in this study was obtained from Kaggle repository which consists of 45 features (attributes) and 541 data points. This dataset was balanced using the Synthetic Minority Oversampling Technique (SMOTE) and features were selected by employing firefly and fruitfly optimization algorithms. The firefly optimized algorithm with Random Forest obtained an accuracy score of 95.205% with 18 selected features. The KNN with firefly algorithm used 13 features and obtained an accuracy of 91.096%. The SVM with firefly algorithm uses 14 features and obtained an accuracy of 93.151%. The fruitfly algorithm with KNN, SVM and RF obtained and accuracy of 86.986%, 90.411% and 93.151% respectively.