A Non-Invasive Approach for Early Polycystic Ovary Syndrome (PCOS) Prediction

Polycystic Ovary Syndrome (PCOS) is an endocrinological disorder that affects women of reproductive age. The abnormal levels of the male hormone in the female body, cystic ovaries, and irregular ovulation are the three most common factors related to PCOS. If this hormonal imbalance goes untreated then the long-term hormonal imbalance can cause infertility, endometrial cancer, type 2 diabetes, and cardiovascular diseases. This research paper proposed a method to predict PCOS with available symptoms using Machine Learning and Deep Learning methods. This method can reduce the complexity of PCOS diagnosis. A Kaggle dataset is used which contains data from 541 women where 177 have PCOS. Chi-square and Extra Tree Classifier methods are used to extract features. Among machine learning models, RFC achieved the highest value accuracy of 86.4% using the Chi-Square feature extraction method. Additionally, Deep Learning models are also analyzed where LSTM achieved the highest accuracy of 89.91% using the Extra Tree classifier feature selection method. A Web Application is developed with non-invasive features that allow any woman to determine whether or not she may have the possibility of having PCOS. This app made diagnosis more widely available.