Automated univariate time series forecasting with regression trees

TL;DR

The paper tackles automatic univariate time series forecasting by leveraging regression trees and their ensembles (bagging and random forests) within an autoregressive, recursive forecasting framework. It proposes practical strategies to handle trending and seasonal behavior, notably an additive transformation for trend-related shifts and season-length-based lag selection, and demonstrates that ensembles can achieve forecast accuracy competitive with exponential smoothing and ARIMA baselines. Through M4-based experiments and extensive demonstrations, the authors show that random forests often provide robust improvements, while simple transformations and appropriate lag choices are crucial for capturing structure in the data. The work is packaged into the utsf R toolkit, offering an automatic, user-friendly workflow for practitioners to generate forecasts without extensive tuning.

Abstract

This paper describes a methodology for automated univariate time series forecasting using regression trees and their ensembles: bagging and random forests. The key aspects that are addressed are: the use of an autoregressive approach and recursive forecasts, how to select the autoregressive features, how to deal with trending series and how to cope with seasonal behavior. Experimental results show a forecast accuracy comparable with well-established statistical models such as exponential smoothing or ARIMA. Furthermore, a publicly available software implementing all the proposed strategies has been developed and is described in the paper.

Figures (17)

• Figure 1: Feature space partition using a regression tree.
• Figure 2: Regression tree associated with feature space subdivision in Fig. \ref{['fig_fsp']}.
• Figure 3: Underfitted and overfitted regression tree models for the example data set.
• Figure 4: An artificial quarterly time series and its forecast.
• Figure 5: An autoregressive tree for the time series in Fig. \ref{['fig_seasonal_series']} based on the previous value of the response.