A Meta-learner Based Ensemble System of Neural Networks for Improving the Accuracy of Preterm Birth Prediction

The birth in which the babies born during 37th, 38th or 39th weeks are considered as Term Births (TB). On the other hand, earlier births are considered as Preterm Births (PTB) and the latter births are considered as Late-term Births or Post-term Births. The risk of neonatal death is very high in PTB. The objective of this study is to introduce an Ensemble System of Neural Networks (ESNN) for improving the accuracy of predicting PTB. This study builds an ensemble system that combines the output of multiple neural networks with different configurations to predict the PTB. A meta learner is used as the combination scheme to arrive at the final output. Choosing a set of suitable base configurations with optimal values set for the hyper-parameters and a suitable meta learner result in improving the accuracy of predicting PTB. The novelty of the proposed method is two folded: (i) Using a Meta Learner rather than simply ensembling neural networks and (ii) Choosing the optimized set of Neural Network as the base learners. Testing is carried out with a dataset that contains 2600 samples with TB as the majority class, the proposed approach has achieved an accuracy of 94%. This system can help the O&G consultants to accurately diagnose the preterm birth and medicate the expectant mother well in advance so that the birth can be delayed, to convert PTB to TB. Rather than training the base learners one after the other, parallel training of Neural Networks with a larger dataset can be considered for future improvements of this study.