LONG SHORT TERM MEMORY (LSTM)-BASED DEEP LEARNING MODEL FOR DETECTING RESPIRATORY DISEASES IN NEONATES

Abstract

Neonatal death remains a major health concern in the world. 45% of under-five deaths are those of neonates, and 40% of neonates’ deaths occur within the first 24 hours. The most deadly of these diseases are those associated with breath. Previous research findings have shown that decrease in world neonatal mortality rate is slower than infant and under five mortality rates, especially in the sub-Saharan African countries because of lack of the state of the art technology for neonatal health care. There is a need to explore the potential of artificial intelligence techniques such as deep learning in this field. This research focused of the development of classification model to predict respiratory diseases in newborns. Data was collected from 1800 hospital records of previously treated neonates in two major hospital in south west Nigeria: Federal Teaching Hospital Ido Ekiti, Ekiti State and Ladoke Akintola University Teaching Hospital Ogbomosho, Oyo Sate Nigeria. The data was preprocessed and use in training two deep learning models; ANN and LSTM. The model was evaluated on precision, recall, F1-score and accuracy. Comparing the performances of the two models during the training and evaluation, LSTM learned well on the training data and gave the best performance on the evaluation metrics with a precision of 83%, a recall of 82%, F1-Score of 80% and an overall system accuracy of 83%. This study shows that LSTM model that gave the best performance is an efficient model and is therefore recommended for detecting respiratory diseases in neonates.