Comparative Analysis of Feature Selection Methods for Automatic Classification of PCOD

Polycystic ovary syndrome (PCOS) has emerged as a prevalent condition affecting a significant proportion of women during their reproductive years across the globe. Presently, timely and accurate diagnosis of PCOS is impeded by a shortage of proficient medical personnel and a large no. of medical tests leading to increased costs and longer time for analysis. To address this challenge, the introduction of Machine Learning (ML) can expedite the task and enable early diagnosis of PCOS. The present work proposes an effective method to automatically classify PCOD-related attributes using different feature selection and ML methods. A comparative analysis and benchmarking of three feature selection methods namely select-K-best, correlation feature selection, and recursive feature elimination, and ten ML methods namely nearest centroid, Bernoulli, support vector machine, LGBM, RFC, ZGB, K-neighbours, bagging, Gaussian and logistic regression has been done with the help of various evaluation metrics and test set analysis. Results indicate that Bernoulli outperformed all the algorithms when employed to features selected by correlation feature selection and achieved an accuracy of up to 92%. In comparison to the existing field, improved evaluation metrics have been achieved.