Predicting congenital heart disease using machine learning techniques

Abstract

A plethora of studies have shown that heart diseases have become the number one killer in the world. Congenital Heart Defect (CHD) is one of the most commonly observed heart defect in an increasing number of newborns. Various research work has been done for CHD detection based on clinical data, however very less work has been with non-clinical data of the patient. Hence in this work, we explore the best suitable approach for non-clinical information, which is prominently no invasive, of the pregnant women in diagnosing the chances of CHD in a child that lives in their womb. For this purpose, a variety of machine learning algorithms including the Naïve Bayes (NB) and the Feedforward Artificial Neural Network (ANN) have been explored. Few relevant researches done previously on the dataset of this problem, used the weighted Support Vector Machine (WSVM) and the weighted Random Forest (WRF) models, however, the state of the art neural network model, used in this paper, significantly improved the CHD detection performance. The results show that the ANN model yields a high accuracy of 99.6% and an impressive weighted accuracy of 90.9% on the selected dataset. Hence it may be strongly suggested to build an ANN-based model for CHD detection in yet to be born babies using such noninvasive nonclinical datasets.

Subject Classification:

• Primary 68T05
• Secondary 98B20

Keywords:

• Congenital Heart Defect
• Pregnancy
• Artificial Neural Network
• weighted Random Forest
• weighted SVM
• Naïve Bayes