An efficient self-organizing map learning algorithm using the set of nearest neurons

The Self-organizing map (SOM) has been extensively applied to image analysis, data clustering, dimension reduction, and so forth. The conventional SOM find the winner neuron and update the weights of winner and its neighborhood regardless of distance from input. In this study, we propose a modified SOM which calculate the distance from input data and find the nearest neuron among neighborhood of winner neuron (BMU). It also calculates the winning frequency of each neuron. We apply modified SOM to various input data set and investigate the performance of both SOM using three standard measurements. We conclude that modified SOM reaches to all input data in better way compare to conventional SOM. The modified SOM preserves the input topology in much better way compare to conventional SOM. The modified SOM self organize in better way than the conventional SOM in every corner of the input data.