Live Births Forecasting Across Health Regions of Goiás Using Artificial Neural Networks: A Clustering Approach.

Predictive models in healthcare can contribute to both clinical practices and administrative decision-making. Anticipating the number of births in a given location may support planning the appropriate service that will be demanded by women during pregnancy, labor, birth, postnatal and newborn care. This study aims to predict the number of live births in the state of Goiás (Brazil), providing a predictive basis to support decision-making for maternal policy strategies. The dataset used to train the proposed model was extracted from the information system on live births of the information department of the single health system (SINASC-DATASUS), containing the number of live births between the years 2000 to 2020 for the health regions of the state of Goiás, Brazil. This proposal employs an approach using Artificial Neural Networks (ANN), demonstrating the comparison of the representation of multivariate time series in the task of clustering birth data by health regions using K-Means and how the combination of these series can improve prediction in ANN-based models. Clusters were formed using a similarity score based on dynamic time warping. It is concluded that the proposed model combining clustering using K-Means and ANN model is a good strategy, generating an average result of 5.5985 and 18.1360 for MAPE and MAE.