Automated Detection of Polycystic Ovary Syndrome Using Machine Learning Techniques

Abstract

Polycystic Ovary Syndrome (PCOS) is a medical condition affecting the female’s reproductive system causing ano/oligoovulation, hyperandrogenism, and/or polycystic ovaries. Due to the complexities in diagnosing this disorder, it was of upmost importance to find a solution to assist physicians with this process. Therefore, in this study, we investigated the possibility of building a model that aims to automate the diagnosis of PCOS using Machine Learning (ML) algorithms and techniques. In this context, a dataset that consisted of 39 features ranging from metabolic, imaging, to hormonal and biochemical parameters for 541 subjects was used. First, we applied pre-processing on the data. Hereafter, a hybrid feature selection approach was implemented to reduce the number of features using filters and wrappers. Different classification algorithms were then trained and evaluated. Based on a thorough analysis, the Support Vector Machine with a Linear kernel (Linear SVM) was chosen, as it performed best among the others in terms of precision (93.665%) as well as high accuracy (91.6%) and recall (80.6%).