An Improved MULTI-SOM Algorithm

Abstract

This paper proposes a clustering algorithm based on the Self Organizing Map (SOM) method. To find the optimal number of clusters, our algorithm uses the Davies Bouldin index which has not been used previously in the multi-SOM. The proposed algorithm is compared to three clustering methods based on five databases. Results show that our algorithm is as performing as concurrent methods.

Self-organizing map of soil properties in the context of hydrological modeling

“…However, an optimal number of clusters are desirable to reduce even more the complexity of soil properties without losing the soil representativeness of the property into the cluster. For this, this algorithm uses the Davies Boulding index for clustering by using an unsupervised classification learning technique to obtain homogeneous partitions of the object while promoting the heterogeneity between partitions [37].…”

“…where variable c defines the number of clusters, i and j denote the clusters, d ( X i ) and d ( X j ) are the distances between all objects in clusters i and j to their respective centroids, and d(c i ,c j ) is the distance between centroids. Smaller values of DB index show better clustering quality [37,39]. Once the DB index is defined for a range of clustering domain, a graph that include the lower mean distance of clustering process is reached, indicates that the minimum compromise of those values is an optimal solution for number of clusters.…”

A Comparative Study of Multi-SOM Algorithms for Determining the Optimal Number of Clusters

“…Reference [14] applied multi-SOM to real data sets to improve multi-SOM algorithm introduced by [2].…”

Hydrological modelling improvements in the assessemnt of water resources of agrarian subbasins in semiarid regions

“…However, an optimal number of clusters are desirable to reduce even more the complexity of soil properties without losing the soil representativeness of the property into the cluster. For this, this algorithm uses the Davies Boulding index for clustering by using an unsupervised classification learning technique to obtain homogeneous partitions of the object while promoting the heterogeneity between partitions (Khanchouch et al, 2013).…”

“…where variable c defines the number of clusters, i and j denote the clusters, ( ) and ( ) are the distances between all objects in clusters i and j to their respective centroids, and ( ) is the distance between centroids. Smaller values of DB index show better clustering quality (Fonseka and Alahakoon, 2010;Khanchouch et al, 2013). Once the DB index is defined for a range of clustering domain, a graph that include the lower mean distance of clustering process is reached, indicates that the minimum compromise of those values is an optimal solution for number of clusters.…”