Self-organizing neural-network-based pattern clustering method with fuzzy outputs

doi:10.1016/0031-3203(94)90148-1

Pattern Recognition

Volume 27, Issue 8, August 1994, Pages 1103-1110

https://doi.org/10.1016/0031-3203(94)90148-1 Get rights and content

Abstract

A pattern clustering method based on the Kohonen feature mapping algorithm and the back-propagation multilayer perceptron is described. The method comprises two phases. First, the Kohonen algorithm and a simple cluster labelling procedure is applied to the training data set to divide it into labelled clusters. The data clusters are then employed to train a three-layer perceptron using the error backpropagation technique. Thus the method is self-organizing by virtue of the Kohonen algorithm and naturally produces fuzzy outputs as a consequence of the backpropagation network. The results of using the proposed method on two standard clustering problems are presented. These show that the method has superior performance compared to crisp clustering networks such as the Kohonen feature map and the ART-2 network.

References (13)

T. Kohonen
Self-organization and Associative Memory
(1984)
P.J. Werbos
Beyond regression: new tools for prediction and analysis in the behavioral sciences
D.B. Parker
Learning logic
D.E. Rumelhart et al.
Learning internal representation by error propagation
G.A. Carpenter et al.
ART-2: Self-organization of stable category recognition codes for analog input patterns
Appl. Optics
(1987)
P. Simpson
Fuzzy adaptive resonance theory

There are more references available in the full text version of this article.

Cited by (17)

Invariant pattern identification by self-organising networks
2001, Pattern Recognition Letters
A new edited k-nearest neighbor rule in the pattern classification problem
2000, Pattern Recognition
A new edited k-nearest neighbor (k–NN) rule is proposed. For every sample y in the edited reference set, all the k- or ( $k+ l$ )-nearest neighbors of y must be in the class to which y belongs. Here $l$ denotes the number of samples which tie with the kth nearest neighbor of y with respect to the distance from y. The performance of the rule proposed has been investigated using three classification examples. As a result, it is shown that the rule proposed will yield good results in many pattern classification problems.
Fuzzy clustering approach in unsupervised sea-ice classification
1999, Neurocomputing
Fuzzy clustering approach for the determination of sea-ice types in polar regions from passive radar images is considered in this paper. The sea-ice concentration in polar regions is generally monitored by multichannel passive microwave imager, and the sea-ice types in polar regions are determined by clustering. However, the boundaries between different sea-ice types are often fuzzy and the results obtained by hard clustering algorithms are often not acceptable. In this paper, the fuzzy clustering algorithm is applied to segment arctic regions by sea-ice types. The number of clusters are determined by partition density method. The experimental results with daily observations for one-year period show that sea-ice types found by fuzzy clustering approach are stable. The formation and growth of sea-ice can also be monitored from the clustered results.
Recursive analysis and forecast of superimposed generation and load time series
2015, International ETG Congress 2015; Die Energiewende - Blueprints for the New Energy Age
A genetic algorithm-based approach for selecting input variables and setting relevant network parameters of a SOM-based classifier
2011, International Journal of Simulation: Systems, Science and Technology
Integration of geographic information system and RADARSAT synthetic aperture radar data using a self-organizing map network as compensation for real-time ground data in automatic image classification
2010, Journal of Applied Remote Sensing

View all citing articles on Scopus

View full text

Self-organizing neural-network-based pattern clustering method with fuzzy outputs

Abstract

Self-organization and Associative Memory

Beyond regression: new tools for prediction and analysis in the behavioral sciences

Learning logic

Learning internal representation by error propagation

ART-2: Self-organization of stable category recognition codes for analog input patterns

Appl. Optics

Fuzzy adaptive resonance theory