Exploratory data analysis with Multi-Layer Growing Self-Organizing Maps

This paper presents an unsupervised hierarchical growing self-organizing neural network for exploring, visualizing and clustering the input space in a hierarchical manner. The Multi-Layer Growing Self-Organization Maps (Multi-Layer GSOM), which is an extension of Growing Self-Organization Maps (GSOM) is capable of alleviating the limitations of both hierarchical agglomerative clustering (HAC) algorithms and hierarchical self-organizing maps (SOMs) due to its dynamic nature. In addition to addressing the above drawbacks, our model is motivated by the hierarchical sensory information processing mechanism in the human brain. The proposed novel algorithm employs several layers of GSOMs (which represent the cortical layers of the human brain) to build cluster (concept) hierarchy for the input data space and subsequently calculates a cluster validity index for each layer to select the best GSOM layer which represent the input space more accurately. The proposed approach was first tested and analyzed using a publicly available bench mark data set. Then we analyzed the performance and strength of the proposed algorithm using gene expression data obtained on Saccharomyces cerevisiae. The results prove that the novel Multi-Layer GSOM consistently improves the accuracy, validity and performance of the cluster hierarchies obtained.