Meta-Heuristic Search Optimization and its application to Time Series Forecasting Model

Abstract

Recently, ensemble learning has been widely used to generate better generalization ability in time series forecasting. The estimation accuracy of the ensemble model depends on the hyper-parameters tuning. The purpose of this research is to introduce a novel kernel-based ensemble machines (KBEM) to determine the best kernel structure with fine-tune hyper-parameters. The optimization method using a hybrid scheme is a combination of meta-heuristic search and local weight learning (LWL) in the process of generating ensemble layers. In order to determine the efficiency of KBEM method support vector regression (SVR) is utilized as a base learner for ensemble. The case studies are based on six benchmarks and global oil consumption data sets. The final output of the proposed model is compared with the current state-of-the-arts tuning techniques, which show that KBEM has achieved obvious advantages in terms of time complexity and performance analysis. Therefore, the KBEM scheme can be regarded as a promising method for automatic kernel regularization and hyper-parameter adjustment of integrated models.

Keywords

• Time series forecasting
• Ensemble learning
• Ensemble tuning
• Hyper-parameters optimization
• Meta-heuristic Search