An Automated Time Series Modeling and Forecasting Approach based on SPSS Statistics
Xinrong Han · 2021
IBM SPSS Statistics provides some traditional models for time series data, including Autoregressive integrated moving average (ARIMA) and exponential smoothing models. These features are accessible via the software's user interface. In order to identify the best-fitting model for a time series, the user needs to go through trial and error. Manually operating the pull-down menus of SPSS Statistics is time-consuming and tedious. Therefore, this paper presents an automated time series modeling and forecasting approach based on the integration plug-in for python in SPSS Statistics, enabling non-professionals to use it for data analysis. Our approach is evaluated on a real-world time series dataset. The results reveal that we can get a satisfactory model outperforming the Expert Modeler provided by SPSS Statistics, and the Mean Absolute Percentage Error (MAPE) on the test data is reduced by about 18.8%. Moreover, we can also get 110 models with better goodness of fit and forecasting performance.