Use of Machine Learning Techniques for Early Prediction of Birth Cases

Preterm birth (PTB) of babies causes high chance of brain-paralysis, breathing-difficulty, low-catch-up growth, and so on. Analyses performed by experts in laboratory for PTB detection are expensive and time consuming. The present work proposes a two-phase model, where Phase-I intends to choose the core parameters to cause PTB based on experts’ opinion and literature review, and develops a machine learning-based graphical user interface (GUI) for early prediction of PTB. In this purpose, 15 socio-demographic parameters are identified from a man-made digitized birth dataset(DD-I), collecting paper-form data from a hospital of Jharkhand. Parameters are statistically tested to check association with PTB. Next, random forest(RF) learner is investigated as the better learner for early prediction, reaching mean accuracy 88.02%(with s.d : 0.001) and more accurate than conventional manual approaches. False positive rate(FPR) of RF is considerably low(7.8%) compared to other experimented learners. In Phase-II, an up-to-date real birth-data collection tool is developed, extending DD-I template.