Combining Forecasts using Meta-Learning: A Comparative Study for Complex Seasonality

In this paper, we investigate meta-learning for combining forecasts generated by models of different types. While typical approaches for combining forecasts involve simple averaging, machine learning techniques enable more sophisticated methods of combining through meta-learning, leading to improved forecasting accuracy. We use linear regression, knearest neighbors, multilayer perceptron, random forest, and long short-term memory as meta-learners. We define global and local meta-learning variants for time series with complex seasonality and compare meta-learners on multiple forecasting problems, demonstrating their superior performance compared to simple averaging.

I. Introduction

Ensemble methods are widely recognized as a cornerstone of modern machine learning (ML) [1], commonly used for regression and classification problems. In addition, ensembling has proven to be a highly effective approach for increasing the predictive power of forecasting models. The ensemble approach in forecasting, which involves combining the predictions of multiple models, can be justified for several reasons. First of all, it usually leads to increased accuracy. Ensemble models often outperform individual models, as they leverage the strengths of different models and minimize their weaknesses. By combining diverse models, the ensemble can produce more accurate predictions by capturing a broader range of patterns and insights from the data. Ensembling also allows for the incorporation of multiple drivers into the data generating process, mitigating uncertainties regarding model form and parameter specification [2]. This helps to reduce the risk of relying on a single model’s limitations or biases, and enables a more comprehensive representation of the underlying data generating process.