Using the hybrid EMD-BPNN model to predict the incidence of HIV in Dalian, Liaoning Province, China, 2004–2018

Background: Acquired immunodeficiency syndrome (AIDS) is a malignant infectious disease with high mortality caused by HIV (human immunodeficiency virus, and up to now there are no curable drugs or effective vaccines. In order to understand AIDS’s development trend, we establish hybrid EMD-BPNN (empirical modal decomposition and Back-propagation artificial neural network model) model to forecast new HIV infection in Dalian and to evaluate model’s performance. Methods: The monthly HIV data series are decomposed by EMD method, and then all decomposition results are used as training and testing data to establish BPNN model, namely BPNN was fitted to each IMF (intrinsic mode function) and residue separately, and the predicted value is the sum of the predicted values from the models. Meanwhile, using yearly HIV data to established ARIMA and using monthly HIV data to established BPNN, and SARIMA (seasonal autoregressive integrated moving average) model to compare the predictive ability with EMD-BPNN model. Results: From 2004 to 2017, 3310 cases of HIV were reported in Dalian, including 101 fatal cases. The monthly HIV data series are decomposed into four relatively stable IMFs and one residue item by EMD, and the residue item showed that the incidence of HIV increases firstly after declining. The mean absolute percentage error value for the EMD-BPNN, BPNN, SARIMA (1,1,2) (0,1,1) in 2018 is 7.80%, 10.79%, 9.48% respectively, and the mean absolute percentage error value for the ARIMA (3,1,0) model in 2017 and 2018 is 8.91%. Conclusions: The EMD-BPNN model was effective and reliable in predicting the incidence of HIV for annual incidence, and the results could furnish a scientific reference for policy makers and health agencies in Dalian.