Accessible Polycystic Ovarian Syndrome Diagnosis Using Machine Learning

Abstract

Polycystic Ovarian Syndrome also known as PCOS is an endocrine syndrome affecting women of reproductive age worldwide. It is one of the leading causes of infertility in women. PCOS is a complex syndrome that also causes insulin resistance in women - when the body can make insulin but can’t use it effectively; which multiplies the risk of developing diabetes. Furthermore, women with PCOS often have higher levels of androgen (male hormones present in females), which can subsequently lead to anovulation (absence or lack of ovulation) and cause irregular periods, acne, thinning scalp hair, and excess hair growth on the face and body. As per the results of a large scale survey conducted across India in 2020, about 16 per cent of women respondents between the ages of 20 and 29 years suffered from polycystic ovary syndrome. PCOS, if left untreated can lead to serious health issues like heart diseases, acne scars, cancer and infertility. Despite the long term effects and fatal implications, the majority of women with PCOS remain undiagnosed. The taboo attached to the diagnosis of the syndrome clubbed with a lengthy initial diagnostic phase can be one of the many contributing factors to the alarming undiagnosed cases. To help bridge the gap of accessibility of PCOS diagnosis, in this paper we demonstrate a web app for women to easily predict their chance of suffering from the syndrome from the comfort of their homes, till help is available to them. We use minimal but optimal clinical features for the prediction of the same, by using feature selection techniques to cherry-pick the parameters which are easily measured at home and are non-invasive. We use the random-forest classifier which has shown promising results in the past for the diagnosis of PCOS using Machine Learning techniques. This paper uses the Kaggle dataset of 541 data points with 41 clinical attributes like weight gain, follicles in the left/right ovary, LHS levels, hair loss, BMI, pimples, etc.