Prognosis of Poly-Cystic Ovary Syndrome and Hormonal Imbalance in Young Women: An Enhanced Feature Selection Approach

PCOS (Poly-Cystic Ovary Syndrome) is an acute medical condition that many women encounter, especially in reproductive age. Obesity, hirsutism, anovulation, pain, hyperandrogenism and oligomenorrhea are typical signs of PCOS. Diagnosing these symptoms is difficult and complicated because they differ greatly among individuals. Utilizing Machine Learning techniques to provide precise and effective diagnostic solutions can be revolutionary in addressing these issues. Large dataset, including as imaging results, hormonal data and patient medical histories, can be effectively analyzed by these algorithms to help diagnose PCOS. In this article, Machine Learning based PCOS prediction model has been developed for the early detection of PCOS. Threshold based Ensemble Filter method has been employed to identify the most pertinent features. A common issue in real-world application is the instability of the feature selection algorithms when it is applied repeatedly to dataset that are similar or with slight variations. In this article, Spearman rank correlation is employed to evaluate the robustness of set of attributes produced by each feature selection methods. Additionally, a number of classifiers have been used, such as Random Forest, SVM (Support Vector Machine), Linear Regression, K-Nearest Neighbour, Naïve Bayes, and perceptron for the PCOS classification. Finally, Random Forest (RF) outperformed well than other models with an accuracy of 96.12% and AUC of 99.63%.