Using Discriminant Analysis to Verify the Clustering of Self-Organizing Map

The data models according to the hot spots spreading in Indonesian forests are usually available with the large of feature space and heterogeneous of distribution patterns. The complexities of this hot spot data structure are central to the present analysis. Clustering of the hot spot regions that persist over time are good indicators of fire risk problems. Therefore, the self-organizing map (SOM) was implemented for clustering hot spot regions. This method is a nonlinear statistical technique that can be used for solving data problems that involved classification and information visualization. The finding of study shows that SOM has provided a classification of hot spot via regions into some different clusters. However, a specification of the cluster is needed when the SOM nodes does not clearly reveal the borders of cluster. Under these circumstances, a supervised learning of discriminant analysis (DA) is used to validate the SOM clusters. The main purpose of DA is to predict cluster membership according to a given prior cluster information, through distance measures and distinct coloring of the nodes in the SOM. DA gave highly accurate cluster discrimination, which shows that this method can be a useful tool to verify the SOM clustering. The combination of the proposed methods is a reliable means of classifying and visualizing of the data, and enables interpretation of the disparities of fire risk by regions in forest on the basis of the hot spot data. Keywords : Cluster validation, complex data, nonlinear method, supervised learning Mathematics Subject Classification: 62P2