Gabriel Ulander Voltaire
To visualize important information and generating a well suited colormap to represent the features of interest in a dataset is complicated. Some colormaps provide with less accurate insights of the data. Meanwhile other colormaps can provide effectively a stronger understanding and identification of features in the data. Therefore, a survey was made to investigate how participants interpet numerical values from colormaps and determining relative distances in the neural network Self-organizing map's grid. We can conclude that using the colormaps in some contexts seemed to work fine. This can be shown from violinplots about the relative distances where some results were quite good. The combination of where the best matching units were located in the grid and choice of colormap can be very effective. However, some cases it might be ineffective due to knowledge of the mechanics in a SOM, familiarity of colormaps or even possible color vision deficiency. Also, when test participants were asked to interpret the underlying distribution in colors as a whole, there were misconceptions. Nonetheless, when used in the context of determining the relative distance was fairly good. There could be implications that maximum and minimum associations were not necessary. Significance testing with the Kruskal-Wallis H-test on the colormaps indicated that there were no difference between the medians of the different colormaps. This meant that one can conclude that the colormaps belong to the same distribution. In this thesis there were 62 test participants. Further investigations to be able to draw other conclusions with high statistical power regarding the different colormaps, perhaps more participants would be required.