Pattern classifying neural network based on Fisher's linear discriminant function
Jong Chan Lee, Yung Hwan Kim, Keon Myung Lee, Suk Hoon Lee · 2003
The network model for two-class pattern classification originally proposed by C. Koutsougeras and C.A. Papachristou (1988) is extended to n-class pattern classification. The proposed model has the advantage of expanding the network by adding the units during the partitioning of the input space while other models have the network topology specified. The result is compared with that of ID3, which is known as a knowledge acquisition tool in machine learning. The comparison shows that the proposed model leads to a better correct rate. This improvement might result from the additional consideration of the optimal projection direction, which ID3 does not consider.>