Meta-learning for time series forecasting and forecast combination

Abstract

In research of time series forecasting, a lot of uncertainty is still related to the task of selecting an appropriate forecasting method for a problem. It is not only the individual algorithms that are available in great quantities; combination approaches have been equally popular in the last decades. Alone the question of whether to choose the most promising individual method or a combination is not straightforward to answer. Usually, expert knowledge is needed to make an informed decision, however, in many cases this is not feasible due to lack of resources like time, money and manpower. This work identifies an extensive feature set describing both the time series and the pool of individual forecasting methods. The applicability of different meta-learning approaches are investigated, first to gain knowledge on which model works best in which situation, later to improve forecasting performance. Results show the superiority of a rankingbased combination of methods over simple model selection approaches.

Load forecasting using a multivariate meta-learning system

“…They have shown that a meta-learning system outperforms approaches representing competition entries in any category. On NN5 competition dataset their Pooling meta-learning had SMAPE of 25.7 which is lower than 26.5 obtained by Structural model, the best performing of 15 single algorithms (Lemke & Gabrys, 2010). If an approach with performance close to or better than the meta-learning system is found, many meta-learning approaches can include those candidates thus becoming better.…”

A mutual association based nonlinear ensemble mechanism for time series forecasting

“…An Elman ANN or EANN is a recurrent neural network that fundamentally differs from the traditional FANN through including an additional context layer and feedback connections [29,30]. During network operations, the outputs of the hidden layer are again fed back to the context layer at each step, so that the context layer can keep track of the previous processing information.…”

“…In this manner, the model with more error receives less weight and vice versa. Ordinary Least Square (OLS) is another popular method that determines the combining weights by minimizing the total Sum of Squared Error (SSE) [5,20,29]. Bunn's outperformance method [10] adopts a Bayesian framework of subjective probabilities and assigns the weight to a model on the basis of the number of times it performed best in the past in-sample forecasting trials.…”

“…This may be primarily due to the lack of well-defined theoretical settings or success records [29]. Few years ago, Freitas and Rodrigues [17] provided some useful insights in this direction and inspired by their concept, Adhikari and Agrawal [1] proposed a nonlinear forecasts combination technique.…”

Towards Automatic Composition of Multicomponent Predictive Systems

“…Hence automatic model selection has been attempted in different ways, e.g. active testing [7], meta-learning [8] and information theory [9]. A common theme in the literature is comparison of different models using data always pre-processed in the same way.…”