A predictive modeling for health expenditure using neural networks strategies

Abstract

The rising cost of healthcare has become a global concern and is a significant challenge for low and middle-income countries, including Jordan. This study aims to develop new neural network models for predicting total health expenditure (EH) in Jordan over the period of 1990–2021. In the first part, the input factors will be selected by examining the relationship between EH and its determinants using various econometric models, including Auto-regressive Distributed Lag Models (ARDL), Error Correction Model (ECM), diagnostic tests, and causality tests. In the second part of the study, we used the selected variables with two neural network models, the Adaptive Neuro-Fuzzy Inference System (ANFIS) and the Hybrid Neural Fuzzy Inference System (HyFIS), to predict the output variable (EH). The input variables are the number of physicians (NM), the number of beds in hospitals (NB), the population size (POP), and the consumer price index (CPI). The findings of this research show that the EH and its determinants exhibit a short and long-run equilibrium link as well as a logical cointegration relationship according to the ARDL boundary test. The results also find that the HyFIS model is more accurate compared to the ANFIS model in predicting total health expenditure. The results also showed that the (HyFIS) model provides the best prediction of health spending compared to the (ANFIS) model based on the error criteria including five measures: mean error (ME), root mean squared error (RMSE), mean absolute error (MAE), mean percentage error (MPE), and mean absolute percentage error (MAPE).