auto-sktime: Automated Time Series Forecasting
Marc-André Zöller, Marius Lindauer, Marco F. Huber
TL;DR
auto-sktime is an Automated Time Series Forecasting framework that leverages AutoML and Bayesian optimization to automatically assemble forecasting pipelines from statistical, ML, and DNN models. It introduces pipeline templates, a warm-starting strategy from prior runs, and multi-fidelity optimization to efficiently search across heterogeneous model families. Evaluations on 64 real-world datasets show competitive or superior forecasting accuracy with minimal human intervention, highlighting practical scalability and robustness across diverse time series. The work delivers a principled, hands-off approach to time series forecasting, supported by a rigorous comparative analysis using MASE, RMSE, and sMAPE metrics across extensive datasets. The combination of automated pipeline construction and careful methodological adaptations advances automated forecasting in real-world settings.
Abstract
In today's data-driven landscape, time series forecasting is pivotal in decision-making across various sectors. Yet, the proliferation of more diverse time series data, coupled with the expanding landscape of available forecasting methods, poses significant challenges for forecasters. To meet the growing demand for efficient forecasting, we introduce auto-sktime, a novel framework for automated time series forecasting. The proposed framework uses the power of automated machine learning (AutoML) techniques to automate the creation of the entire forecasting pipeline. The framework employs Bayesian optimization, to automatically construct pipelines from statistical, machine learning (ML) and deep neural network (DNN) models. Furthermore, we propose three essential improvements to adapt AutoML to time series data. First, pipeline templates to account for the different supported forecasting models. Second, a novel warm-starting technique to start the optimization from prior optimization runs. Third, we adapt multi-fidelity optimizations to make them applicable to a search space containing statistical, ML and DNN models. Experimental results on 64 diverse real-world time series datasets demonstrate the effectiveness and efficiency of the framework, outperforming traditional methods while requiring minimal human involvement.