Data Mining and Risk Analysis Supporting Decision in Brazilian Public Health Systems

Health data monitoring is a crucial activity to reduce maternal, neonatal and infant mortality rates. Available data in Brazilian health databases point that It is possible to predict death risk in the early stages of gestation and infant development. In this research, we consider the information availability still in the gestational period to propose different death risk prediction models for this public of interest. We also detail the data mining process to apply machine learning-based techniques in death risk classification for maternal, neonatal and infant patients. We present an experiment pipeline to estimate average performance and evaluated machine learning models with different features combinations. Additionally, we show a web service which provides multiple predictive models by information availability. Results show Random Forest obtaining better performance when compared to the other machine learning methods.