SYMPTOM-BASED PREDICTIVE MODEL FOR ENDOMETRIOSIS USING MACHINE LEARNING METHODS

Abstract

Endometriosis is a benign disorder of the female reproductive system. Endometriosis is hard to diagnose and often diagnosed very late, resulting in a poor quality of life and infertility. The only reliable way to know is via laparoscopy. Unfortunately, because of a general lack of knowledge, there is usually a long delay between a woman first going to the doctor regarding her symptoms and being diagnosed The objective of this study is to establish if clinical symptoms can help in diagnosing endometriosis. We apply a combination of machine learning methods, such as lasso, random forests and boosting for endometriosis prediction accompanied by feature selection. The results were verified using cross-validation and unsupervised learning. The study shows that using machine learning the models predict all stage endometriosis with good accuracy and support the symptoms which are highly predictive of endometriosis by experts. However, we cannot generalize our results due to limitations in data. Therefore, it is encouraged that more data is collected.