Self-organizing maps: a novel approach to identify and map business clusters

Abstract

Business cluster identification is an essential topic for helping understand regional and global supply chains and establishing economic policies and logistics. This work aims to leverage the benefits of self-organizing maps (SOM), combined with traditional clustering algorithms and image processing techniques, to identify business clusters that are described by high-dimensionality feature vectors. It is advantageous over previous work because the algorithm is unsupervised and makes no assumptions about the number of clusters for a given feature set. The proposed algorithm was evaluated using recent datasets for US metropolitan cities from the Indiana Business Research Center (Innovation 2.0) and the Occupational Employment Statistics Survey. Data involving innovation metrics, education levels, economic well-being, connectivity, local GDP, and STEM are aggregated to demonstrate the effectiveness of the proposed neural network. The clustering results are compared to traditional approaches, including K-means clustering, both quantitatively and qualitatively. The unsupervised nature of the proposed SOM approach, and the acceptable computational complexity of the overall algorithm, suggests that self-organizing maps offer several advantages over traditional methods. In this work, we present a novel architecture coupling a SOM model with processing techniques for automatically identifying business clusters derived from high-dimensionality feature vectors, the first use case of SOMs in business cases affecting supply chains and other economic decisions. Preliminary results confirm the viability of architecture as an unsupervised approach for identifying business clusters.