The Precision of Neonatal Birth Outcomes Prediction Using the Bagging Neural Network

Abstract

Background and Objectives: The high rate of neonatal mortality is a major problem in health care systems all around the world. The accurate estimation of neonatal mortality is a prerequisite for the development of future health strategies that leads to the improvements in neonatal health. Providing a predictive model is, therefore, essential to reduce the neonatal mortality rate and reducing health care costs. The purpose of this study was to produce a model based on the data mining techniques to increase the accuracy of the prediction of the outcome of the neonatal mortality using a bagging neural network model in Rapidminer software. Material and Methods: This study was conducted on 8053 births (including 1605 cases and 6448 controls) across the country in 1394. The study variables including maternal diseases, mother age, gestational age, child gender, birth weight, birth order, abnormalities were selected as predictive factors for bagging neural network method. We compared bagging neural network with neural network, decision tree and nearest neighbor. Some criteria including the area under ROC curve, precision, accuracy and classification error rate were considered in comparing with other data mining models. Results: The comparison of bagging neural network with other data mining models showed that the bagging neural network gives better results compared to other models: precision (99.21), accuracy (99.17), classification error rate (0.83) and AUC value (0.992). Conclusion: We conclude that the bagging neural network may help to reduce the cost of health care system, and to improve the community health by preventing the mortality and adverse outcomes in neonates.   

Keywords

• neonatal mortality
• data mining
• bagging neural network
• decision tree
• nearest neighbor