Time Series Forecasting to Improve Predictive Modelling in Public Maternal Healthcare Data

Abstract

Background: To predict the future health situation of a nation or a state, the Time Series Predictive Modelling is a valuable tool in the health system. In order to attain the Millennium Development Goals (MDGs) for MMR in India, the purpose of this research work is to identify the trends in the essential components, which are responsible for maternal death prevention. Methods: To achieve the above-mentioned objective we have evaluated the performance of three different approaches in our process model for Time Series Predictive Modelling on public maternal health data. The first approach is Exponential Smoothing method i.e. is a statistics based method, the second one is Multi Layer Perceptron method i.e. a machine learning based method and the third one is Long Short Term Memory network i.e. a deep learning method. For the data analysis, the five years’ monthly time series data (2012 to 2017) of Uttar Pradesh state of India is collected from NRHM portal of Indian Government. It is partitioned into training data for modeling and testing data for validation of the model. Results: The major components from the original data are selected by using an attribute selection method i.e. greedy approach based Best-First Wrapper method. The MAPE statistical parameter is used to define the accuracy level of the predictive values for the selected dimensions of the given data. A performance-based comparison of all applied approaches is presented at last which illustrates that exponential smoothing method has performed better than the other two methods. Conclusion: The presented trend analysis and future values generated by using the presented process model will feed input in the decision making for planning better healthcare services.