Identification of noise outliers in clustering by a fuzzy neural network

Since most real data sets encountered in cluster analysis are contaminated with background noise or outliers, it is essential to detect and isolate these noise samples from the data set. A few noisy points can affect the clustering procedure by severely biasing the algorithm. An ideal solution to this problem is to identify all the outliers and form a separate noise cluster. To do so, a technique is required by which the noise points are automatically identified and removed from the pattern data. The authors present a modified adaptive fuzzy leader clustering (AFLC) algorithm that has been used to detect and eliminate the outliers from the data structure and create a separate cluster of the outliers. The AFLC algorithm has an adaptive resonance theory (ART) like architecture with fuzzy learning embedded into it. Test results of the algorithm when applied to real and noisy data sets are presented