Packed hyper-ellipsoid classifiers
G.D. Tattersall, Yi Kang · Electronics Letters · 1994
Pattern classification is frequently performed using the k-nn algorithm or a neural network. The choice of parameters for the former is often difficult and the amount of data which has to be stored in the classifier can be high. Neural network classifiers can overcome some of these problems but learning is often unreliable and slow. An alternative which combines some of the best features of the k-nn and neural network classifier is decribed by the authors. The classifier is called a packed hyper-ellipsoid classifier.