MULP: A Multi-Layer Perceptron Application to Long-Term, Out-of-Sample Time Series Prediction

Abstract. A forecasting approach based on Multi-Layer Perceptron (MLP) Artificial Neural Networks (named by the authors MULP) is proposed for the NN5 111 time series long-term, out of sample forecasting competition. This approach follows a direct prediction strategy and is completely automatic. It has been chosen after having been compared with other regression methods (as for example Support Vector Machines (SVMs)) and with a recursive approach to prediction. Good results have also been obtained using the ANNs forecaster together with a dimensional reduction of the input features space performed through a Principal Component Analysis (PCA) and a proper information theory based backward selection algorithm. Using this methodology we took the 10th place among the best 50% scorers in the final results table of the NN5 competition.

• Type: Journal
• Year: 2010
• Where: ISNN
• Authors: Eros Pasero, Giovanni Raimondo, Suela Ruffa
• Keywords: Artificial Neural Networks, Direct Prediction Strategy, Example Support Vector, ISNN 2010, Neural Networks