A comparative study of neural network based feature extraction paradigms

Boaz Lerner*, Hugo Guterman#, Mayer Aladjem#, and Its’hak Dinstein# *University of Cambridge Computer Laboratory, New Museums Site, Cambridge CB2 3QG, UK #Department of Electrical and Computer Engineering, Ben-Gurion University of the Negev, Beer-Sheva 84105, Israel Published in Pattern Recognition Letters, vol. 20(1), pp. 7-14, 1999. Abstract The projection maps and derived classification accuracies of a neural network (NN) implementation of Sammon’s mapping, an auto-associative NN (AANN) and a multilayer perceptron (MLP) feature extractor are compared with those of the conventional principal component analysis (PCA). Tested on five real-world databases, the MLP provides the highest classification accuracy at the cost of deforming the data structure, whereas the linear models preserve the structure but usually with inferior accuracy.