Prediction Models for Polycystic Ovary Syndrome Using Data Mining

Abstract

Polycystic Ovary Syndrome is an endocrine abnormality that occurs in the female reproductive system and is considered a heterogeneous disorder because of the different criteria used for its diagnosis. Early detection and treatment are critical factors to reduce the risk of long-term complications, such as type 2 diabetes and heart disease. With the vast amount of data being collected daily in healthcare environments, it is possible to build Decision Support Systems using Data Mining and Machine Learning. Currently, healthcare systems have advanced skills like Artificial Intelligence, Machine Learning and Data Mining to offer intelligent and expert healthcare services. The use of efficient Data Mining techniques is able to reveal and extract hidden information from clinical and laboratory patient data, which can be helpful to assist doctors in maximizing the accuracy of the diagnosis. In this sense, this paper aims to predict, using the classification techniques and the CRISP-DM methodology, the presence of Polycystic Ovary Syndrome. This paper compares the performance of multiple algorithms, namely, Support Vector Machines, Multilayer Perceptron Neural Network, Random Forest, Logistic Regression and Gaussian Naïve Bayes. In the end, it was found that Random Forest provides the best classification, and the use of data sampling techniques also improves the results, allowing to achieve a sensitivity of 0.94, an accuracy of 0.95, a precision of 0.96 and a specificity of 0.96.