i-HOPE: Detection And Prediction System For Polycystic Ovary Syndrome (PCOS) Using Machine Learning Techniques

The present world women population is widely affected by preterm abortions, infertility, anovulation etc. It is observed that polycystic ovary syndrome (PCOS), a condition seen among the women of reproductive age is having a major influence in the cause of infertility. Over five million women worldwide in their reproductive age PCOS. It is an endocrine disorder characterized by changes in the female hormone levels and the abnormal production of male hormones. This condition leads to ovarian dysfunction with increased risk of miscarriage and infertility. The symptoms of PCOS include obesity, irregular menstrual cycle, and excessive production of male hormone, acne, and hirsutism. It is extremely difficult to diagnose PCOS due to the heterogeneity of symptoms associated and the presence of a varying number of associated gynecological disorders. The time and cost involved in innumerous clinical tests and ovary scanning has become a burden to the patients with PCOS. To address this problem this paper proposes system for the early detection and prediction of PCOS from an optimal and minimal but promising clinical and metabolic parameters, which act as an early marker for this disease. The data sets required for this system development are obtained through patient survey of 541 women during doctor consultations and clinical examinations. Out of the 23 features from clinical and metabolic test results, 8 potential features are identified using SPSS V 22.0 based on their significance. Classification of PCOS with the feature set transformed with Principal Component Analysis (PCA) is done using various machine learning techniques such as Naïve Bayes classifier method, logistic regression, K-Nearest neighbor (KNN), Classification and Regression Trees (CART), Random Forest Classifier, Support Vector Machine (SVM) in Spyder Python IDE. Results revealed that the most suitable and accurate method for the PCOS prediction is RFC with an accuracy of 89.02%.