An optimally configured hybrid model for healthcare time series prediction

Abstract

The challenge in improving accuracy mtime series prediction has motivated researchers to develop more efficient prediction models. Prediction of healthcare data such as mortality and morbidity assumes importance in healthcare management as these data serve as health indicators of a society. The accuracy rates obtained using linear models such as autoregressive integrated moving average and linear regression are not high as they have limitations in handling the non-linear relationships among the data. Neural network models are considered to be better in handling such non-linear relationships. Healthcare time series data consist of complex linear and nonlinear patterns and it may be difficult to obtain high prediction accuracy rates using only linear or neural network models. The researchers propose a hybrid method which combines the best linear model with an optimally configured neural network. Unlike other hybrid models which use a predetermined configuration for the linear and neural network components, the proposed method selects the best linear model and the optimum neural network configuration based on the type of input data. The proposed method is tested based on two types of healthcare data, namely infant mortality rate and morbidity of malaria data. Experiment results show that the proposed hybrid model yields more accurate prediction results compared to the other known models. © Medwell Journals, 2011.