A Wrapper for Projection Pursuit Learning

– Constructive algorithms are effective methods for designing Artificial Neural Networks (ANN) with good accuracy and generalization capability, yet with parsimonious network structures. Projection Pursuit Learning (PPL) has demonstrated great flexibility and effectiveness in performing this task, though presenting some difficulties in the search for appropriate projection directions in input spaces with high dimensionality. This characteristic hinders the use of PPL at the field of time series prediction, due to the common occurrence of high-dimensional input spaces. We propose a method based on the wrapper methodology to perform variable selection, so that only a subset of highly informative lags is going to be considered as the regression vector. The results show that variable selection increases the performance of the final ANN at an acceptable increase of computational cost. Keywords – Artificial Neural Networks, Constructive Algorithms, Projection Pursuit Learning, Variable S...

• Type: Conference
• Year: 2007
• Where: IJCNN
• Authors: Leonardo M. Holschuh, Clodoaldo Ap. M. Lima, Fernando J. Von Zuben
• Keywords: Artificial Neural Networks, IJCNN 2007, Projection Pursuit, Variable Selection