Risk Analysis Forecasting Models of Poisson Regression, Negative Binomial Regression, Poisson GSARIMA, and Negative Binomial GSARIMA (Case Study: Number of Bicycle Sales)

Abstract

The Poisson model is a model that can be applied to count data, where in this research the case study used is the number of bicycle sales. However, there is an equidispersion assumption in the Poisson model, that the response variable has the same mean and variance. A more flexible model is needed if the equidispersion assumption is not met, namely the Negative Binomial model. In this research, two models were applied, namely the regression model and the GSARIMA model, with two different distributions, namely the Poisson distribution and the Negative Binomial distribution. Therefore the models that will be compared are the Poisson Regression, Negative Binomial Regression, Poisson GSARIMA, and Negative Binomial GSARIMA models. The differences in results for each model are due to errors that occur in each model used. Hence, a model with a smaller error can be said to be a model that has a smaller risk than other models. The results of this study show that the error rate in the Negative Binomial GSARIMA ZQ1 model is relatively smaller than other models with a value of AIC = 1058.7. This model is the best model that can be used as a forecasting model in the case of bicycle sales and can minimize the risk of error in a forecasting result.