Multi-Verse Optimization of Multilayer Perceptrons (MV-MLPs) for Efficient Modeling and Forecasting of Crude Oil Prices Data

Abstract

Arbitrary changes in crude oil prices makes its forecasting quite difficult. Multilayer neural networks are found to be effective in predicting such cured oil prices. Crafting optimal neural network architecture requires numerous trial and error methods. This article presents a hybrid model based on multi-verse optimization (MVO) of multilayer perceptron (MLP), termed as (MV-MLP), where a universe/individual of MVO represents a potential MLP in the universe of discourse. A set of such universes forms a population and the best universe, i.e. optimal MLP is selected through a search process. The search process starts with a random population, gradually moves toward the global optimum and the optimal MLP is obtained at fly rather fixing it earlier. The proposed MV-MLP is evaluated on forecasting the crude oil prices and the predictability performance is established through comparative study with other trained models trained. Experimental results and comparative study suggests the superiority of MV-MLP based forecasting.