Rule induction for forecasting method selection: Meta-learning the characteristics of univariate time series

For univariate forecasting, there are various statistical models and computational algorithms available. In real-world exercises, too many choices can create difficulties in selecting the most appropriate technique, especially for users lacking sufficient knowledge of forecasting. This paper provides evidence, in the form of an empirical study on forecasting accuracy, to show that there is no best single method that can perform well for any given forecasting situation. This study focuses on rule induction for forecasting method selection by understanding the nature of historical forecasting data. A novel approach for selecting a forecasting method for univariate time series based on measurable data characteristics is presented that combines elements of datamining, meta-learning, clustering, classification and statistical measurement. Over 300 datasets are selected for the empirical study from diverse fields. Four popular forecasting methods are used in this study to demonstrate prototype knowledge rules. In order to provide a rich portrait of the global characteristics of the time series, we measure: trend, seasonality, periodicity, serial correlation, skewness, kurtosis, non-linearity, self-similarity, and chaos. The derived rules for selecting the most suitable forecasting method based on these novel characteristic measures can provide references and recommendations for forecasters.