Birth time prediction based on uterus-activity using machine learning

Abstract

The goal of this paper is to describe a predictive model that is able to estimate the expected time of the child’s birth using contraction data collected since the beginning of labor. During research both classical and neural network time series forecasting models were investigated. Among the classic time series forecasting methods, the Integrated Autoregressive Moving Average Model, i.e., ARIMA, and Holt’s exponential smoothing were examined. And among the time series forecasting methods based on neural networks, the LSTM i.e., Long short-term memory and the one-dimensional convolutional neural network were implemented. The evaluation results show that the neural networks outperformed the classical methods. The best result was achieved using the one-dimensional convolutional neural network.