Applying two-stage SOM-based clustering approaches to industrial data analysis

Abstract

Data analysis is a promising method for reducing the complexity of information management when handling huge amounts of data. In this paper the performances of hybrid two-stage methods combining self-organising map (SOM) and traditional clustering algorithms are presented and evaluated with the goal of identifying the techniques leading to the best clustering quality. The SOM-based two-stage methods are also compared to single-stage approaches applying traditional hierarchical and partitioning algorithms. These comparisons are initially based on the analysis of two reference data sets (Iris and Abalone) which shows how the use of SOM improves clustering quality while reducing computational time. In order to further evaluate the proposed two-stage method, the comparison is extended to two industrial applications. The first one concerns the group technology problem and the second is related to the classification of purchased components. The obtained results show that SOM + K-means can achieve a satisfying clustering assignment quality while reducing the computational time.