In this paper a complex classification scheme including a combination of a cluster generator based on a neural network and three different types of classifiers is proposed. The first part of the scheme consists of a self organized memory S.O.M.-type unsupervised neural network which converges quickly to the clusters' vectors. In the second part, which is the main classification part, three classifiers are being compared. This comparison is made in a such a way so as to study the improvement of the performance and classification when getting from purely classical classification schemes to fuzzy ones and thus emphasize on the usefulness and reliability of the fuzzy set theory. The overall approach aims at showing that neural networks and fuzzy classifiers can be combined in a such a way which exploits the advantages of both approaches in classification problems. Neural networks. on tire one hand, operate very quickly after they are trained but they are not capable of recognizing easily information not familiar to them. On the other hand fuzzy systems overcome with success this drawback because they are generalized by nature. However they do not provide us with straight decisions. Instead, they give an estimation of the nature of the problem associating. They thus give us more than one possible solutions. In literature, the most common way of putting together these two concepts of neural networks and fuzzy systems is the adoption of the neuro-fuzzy schemes which in then turn are constructed by fuzzy neurons instead of classical neurons. This paper proposes a two stage scheme. and in addition studies three classical algorithms a classical one such as Nearer Neighbor N.N.R. a generalization of it. die Fuzzy-N.N.R., and a well known fuzzy classifier die Fuzzy C-Means F.C.M. Results show Üiat fuzzy N.N.R. operates with significantly better performance in terms of classification performance like mean classification error, and operational time expressed by convergence tune. However it is shown Üiat the fuzzy schemes are more reliable than the fuzzified classical schemes. The structure of the scheme is simple. There is no need for long convergence delays and complex learning procedures as the scheme is a S.O.M.-type one. It is especially designed for optimal and quick convergence to the vectors of clusters.
[1]
B. Ripley,et al.
Pattern Recognition
,
1968,
Nature.
[2]
James C. Bezdek,et al.
Efficient Implementation of the Fuzzy c-Means Clustering Algorithms
,
1986,
IEEE Transactions on Pattern Analysis and Machine Intelligence.
[3]
Hon Keung Kwan,et al.
A fuzzy neural network and its application to pattern recognition
,
1994,
IEEE Trans. Fuzzy Syst..
[4]
Paul D. Gader,et al.
A fuzzy logic system for the detection and recognition of handwritten street numbers
,
1995,
IEEE Trans. Fuzzy Syst..
[5]
F. Palmieri,et al.
Hebbian learning and self-association in nonlinear neural networks
,
1994,
Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN'94).
[6]
James C. Bezdek,et al.
On cluster validity for the fuzzy c-means model
,
1995,
IEEE Trans. Fuzzy Syst..
[7]
James C. Bezdek,et al.
Local convergence of the fuzzy c-Means algorithms
,
1986,
Pattern Recognit..
[8]
Laurence Tianruo Yang,et al.
Fuzzy Logic with Engineering Applications
,
1999
.
[9]
Zheru Chi,et al.
Handwritten numeral recognition using self-organizing maps and fuzzy rules
,
1995,
Pattern Recognit..
[10]
Stephen Coombes,et al.
Using generalized principal component analysis to achieve associative memory in a Hopfield net
,
1994
.