Abstract The debate on significance of numerous political, economic and financial indicators driving crude oil prices is perpetual. There is no single indicator which can provide a complete picture of how prices can be determined. Nor a simple combination of input indicators can provide accurate and robust price forecast methods. In particular, feature selection plays a key role in designing a forecasting model for oil prices. However, all existing method of predicting oil prices have accounted for non-linearity, non-stationarity and time-varying structure of crude oil prices but seldom focus on selecting significant features with high predicting power. Besides, there is lack of competent feature selection techniques based on associations and dependency of indicators for designing the input vector of oil price forecast. For this purpose, a novel two-stage feature selection method “MI 3 Algorithm” is proposed for inferring non-linear dependence between oil prices and strategic indicators driving them by employing interaction information and mutual information as measure of redundancy(or synergy) and relevance. The study targets to figure out the importance and impacting mechanism of key indicators driving crude oil prices based on the proposed feature selection algorithm employing multi-layered perceptron neural network (MLP), general regression neural network (GRNN) and cascaded neural network (CNN) as forecasting engine for oil price prediction. The results confirmed the superiority of proposed algorithm compared to some other methods. Besides its high accuracy, the proposed algorithm provides non-redundant and most relevant features as compared to other methods employed in study.
[1]
C. Aloui,et al.
Crude oil price forecasting: Experimental evidence from wavelet decomposition and neural network modeling
,
2012
.
[2]
Nima Amjady,et al.
Design of input vector for day-ahead price forecasting of electricity markets
,
2009,
Expert Syst. Appl..
[3]
Luiz Fernando Loureiro Legey,et al.
Forecasting oil price trends using wavelets and hidden Markov models
,
2010
.
[4]
Heping Pan,et al.
Forecasting model for crude oil prices based on artificial neural networks
,
2008,
2008 International Conference on Intelligent Sensors, Sensor Networks and Information Processing.