Application of ARIMA and SVM combination model in prediction of COVID-19

ABSTRACT

OBJECTIVE: To establish a combination model based on differential autoregressive moving average model (ARIMA) and introduction of support vector machine (SVM) method so as to predict the incidence trend of COVID-19. METHODS: ARIMA model was employed to predict the linear part of data of daily increased confirmed cases and capture the linear change trend of time series data. SVM was employed to predict the non-linear change trends. The mean absolute error (MAE), mean square error (MSE) and mean absolute percentage error (MAPE) were compared between the two combination models so as to assess the prediction results. RESULTS: During fitting stage of the models, as compared with the single ARIMA model and SVM model, the MSE of the ARIMA-SVM combination model was 0.004 (the least) in prediction of COVID-19, the MAE 0.055, the MAPE 0.004. During the prediction stage of the models, the MSE, MAE and MAPE of the combination model were respectively 7.811, 2.730 and 0.764, which were also the least among the 3 models. CONCLUSION: The ARIMA-SVM combination model is more precise than the single ARIM or SVM in prediction of trend of COVID-19.