Identification of nuclear transients via optimized fuzzy clustering

In this paper, we look into the issue of using cluster analysis for transient classification in nuclear components and systems. In general, the choice of the metrics upon which clustering is based can be critical for obtaining geometric clusters as close as possible to the real physical classes in the feature space. The complexity and variety of cluster shapes and dimensions which can be expected in the transient classification of interest lead us to take an approach based on a different Mahalanobis metric for each cluster. The a priori known information regarding the true classes to which the patterns belong is exploited to select, by means of a supervised evolutionary algorithm, the different optimal Mahalanobis metrics. Further, the diagonal elements of the matrices defining the metrics can be taken as measures of the relevance of the features employed for the classification of the different patterns. The efficiency of the approach is verified with respect to a literature problem and then applied to the case of classification of transients in a nuclear component.