3.5. Simulation results
Kohonen׳s SOM neural network performance on the Text or Alphanumeric symbols data set is assessed through three different options: a study reported in the literature [46], in-house simulation using the MATLAB SOM toolbox, and another in-house simulation using the PROWLER (also titled as WSN–SOM).
3.5.1. Solution reported in literature
A solution using the SOM neural network for the text dataset was reported in [46]. The SOM network employed 70 neurons which were arranged in a topology of rectangle to map the five-dimensional data vectors in Table 4 to two dimensions. The SOM solution reported in [46] is re-created and presented in Fig. 6 and Fig. 7. The study did not indicate any quantization or topographical error values for the solution that was reportedly obtained after 10,000 training steps. The authors stated that the calibration was performed by supervised labeling of neurons in response to a specific known pattern vector from the training set.