The Effect of Model Type on the Accuracy of Predicted Values

Abstract

The purpose of this experiment was to determine which model would be the most effective in predicting disease progression given two months of data. If which model is more accurate is known, governments can develop and use the models to respond to disease more effectively. The SIR and SEIR are commonly used compartmental models for diseases. The SIR model incorporates another factor. It was hypothesized the values predicted by the SEIR model would be more accurate than the values predicted by the SIR model. The experiment was conducted by first calculating an SEIR model and an SIR model based on data from the first 2 months of 3 different Ebola outbreaks in 2014. The actual values of data were compared against the predicted values of data by the SIR and SEIR value by use of percent error. The SEIR model had a lower mean percent error than the SIR model, which supports the hypothesis. A t-test performed on the data revealed it was not significant. The null hypothesis of no difference in the accuracy of the data predicted by the SIR model and the data predicted by the SEIR model failed to be rejected. It is believed there is no difference in the accuracy of predictions of the SIR model vs the SEIR model. This may be because there are outside factors such as peoples responses and non-disease related death and births. Research can be done into more adaptable models that are able to accommodate birth and death rates and effects of the responses of people.