Feature-based forecasting algorithms for large collections of time series

Talk given at ACEMS workshop on “Statistical Methods for the Analysis of High-Dimensional and Massive Data Sets” I will discuss two algorithms used in forecasting large collections of diverse time series. Each of these algorithms uses a meta-learning approach with vectors of features computed from time series to guide the way the forecasts are computed. In FFORMS (Feature-based FORecast Model Selection), we use a random forest classifier to identify the best forecasting method using only time series features.<img src=“http://feeds.feedburner.com/~r/ProfessorRobJHyndman/~4/uohe4ANFd …