Abstract
This paper aims to provide combination of multiple prediction models using different strategies including ensemble selection, voting, stacking and multi-schemes to design a model capable of predicting oil prices accurately. Daily data from 1999 to 2012 with 14 variables were used, which were further divided into 10 sub-datasets according to various attribute selection methods. Four groups of training and testing were examined. Experimental results conclude that performance of the combination model works better than author's previous work and ensemble selection outperforms other combination methods.