Effective Polycystic Ovary Syndrome (PCOS) Classification with Machine Learning and Optimized Feature Selection

Abstract

Polycystic Ovary Syndrome (PCOS) is a prevalent endocrine disorder affecting women of reproductive age, necessitating accurate prediction and diagnosis for effective patient care and personalized treatment. Despite numerous studies on PCOS, there remains a gap in the literature regarding the most effective classification algorithms and feature selection methods tailored to PCOS datasets, as well as a comparative analysis of data science tools like Python and RapidMiner in this context. This study aims to address these gaps by identifying the most effective machine learning classification methods for predicting PCOS, determining the significant features involved, and comparing the performance of Python and RapidMiner tools. Key steps include data preprocessing, constructing and evaluating a stacking classifier using Python, and implementing various K-fold cross-validation settings to assess training and test accuracies, precision, and recall. The research examines variables of infertility and non-infertility within the dataset, using a Stacking Classifier model with 37 features to report on cross-validation results. Findings contribute to advancing PCOS prediction by leveraging machine learning techniques and provide insights into the efficiency of Python and RapidMiner in healthcare data science. These outcomes have implications for customizing healthcare, enhancing patient care, and optimizing operational efficiency, ultimately aiming to improve PCOS management and treatment through advanced data science methodologies.