A Boosting Classification Approach Based on SOM

  • Hoa Dinh Nguyen PTIT
Keywords: Self organizing map, Adaptive boosting, Learning vector quantization


Self-organizing map (SOM) is well known for its ability to visualize and reduce the dimension of the data. It has been a useful unsupervised tool for clustering problems for years. In this paper, a new classification framework based on SOM is introduced. In this approach, SOM is combined with the learning vector quantization (LVQ) to form a modified version of the SOM classifier, SOM-LVQ. The classification system is improved by applying an adaptive boosting algorithm with base learners are SOM-LVQ classifiers.  Two decision fusion strategies are adopted in Adaboost algorithm, which are majority voting and weighted voting. Experimental results based on a real dataset show that the newly proposed classification approach for SOM outperforms traditional supervised SOM. The results also suggest that this model can be applicable in real classification problems.