A novel evolutionary multi-objective ensemble learning approach for forecasting currency exchange rates

Due to the potential impact of the (currency) exchange rate risk in the financial market, forecasting exchange rate (FET) has become a hot topic in both academic and practical worlds. For many years, the various methods have been proposed and used for FET problems including the method of the artific...

Full description

Saved in:
Bibliographic Details
Published in:Data & knowledge engineering 2018-03, Vol.114, p.40-66
Main Authors: Bui, Lam Thu, Truong Vu, Van, Huong Dinh, Thi Thu
Format: Article
Language:English
Subjects:
Currency exchange rates forecasting
Ensemble learning
Multi-objective evolutionary
Non-dominated differential evolution
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Due to the potential impact of the (currency) exchange rate risk in the financial market, forecasting exchange rate (FET) has become a hot topic in both academic and practical worlds. For many years, the various methods have been proposed and used for FET problems including the method of the artificial neural network (ANN). However, in many cases of FET, there is the limitation of using separate methods since they are not able to fully capture financial characteristics. Recently, more researchers have been beginning to pay attention to FET based on an ensemble of forecasting models (in other words, the combination of individual methods). Previous studies of ensemble methods have shown that the performance of an ensemble depends on two key elements (1) The individual performance and (2) diversity degree of base learners. The main idea behind this paper comes from these key elements, the authors use ANNs as the base method (or weak learners), and weights of these ANNs will be optimized by using multi-objective evolutionary algorithms (MOEAs) including the Non-Dominated Sorting Genetic Algorithm II (NSGA-II) and the Non-Dominated Sorting Differential Evolution (NSDE) using directional information. To assist MOEAs, a number of diversity-preservation mechanisms are used to generate diverse sets of base classifiers and finally we propose to use modified Adaboost algorithms to combine the results of weak learners for overall forecasts. The results show that the proposed novel ensemble learning approach can achieve higher forecasting performance than those of individual ones.
ISSN:0169-023X
1872-6933
DOI:10.1016/j.datak.2017.07.001