Adaptive weighted-function models for time series prediction

Abstract

Time series prediction has been widely used in various fields. GEP is one of the popular methods for time series analysis. However, the GEP-based prediction models contain only one single function. To accurately capture the dynamic behavior of time series, this study develops a system which integrates multiple functions in a GEP-based model for time series prediction. The weight of each function is determined by the accuracy of its last prediction. In addition, a light local search is applied to adjust the function weights. The experimental results show that the proposed system outperforms several GEP-based approaches.