Abstract
The problem of predicting oil prices is worthy of attention. As oil represents the backbone of the world economy, the goal of this paper is to design a model, which is more accurate. We modeled the prediction process comprising of three steps: feature selection, data partitioning and analyzing the prediction models. Six prediction models namely: Multi-Layered Perceptron (MLP), Sequential Minimal Optimization for regression (SMOreg), Isotonic Regression, Multilayer Perceptron Regressor (MLP Regressor), Extra-Tree and Reduced Error Pruning Tree (REPtree). These prediction models were selected and tested after experimenting with other several most widely used prediction models. The comparison of these six algorithms with previous work is presented based on Root mean squared error (RMSE) to find out the best suitable algorithm. Further, two meta schemes namely Bagging and Random subspace are adopted and compared with previous algorithms using Mean squared error (MSE) to evaluate performance. Experimental evidence illustrate that the random subspace scheme outperforms most of the existing techniques.