Using the LogSig-function to integrate expert knowledge to self-organizing map (SOM) based analysis
Kimmo Hätönen, Sampsa Laine, Timo Similä · 2003
Pre-treatment of data is an important step of data analysis. Most of the analysis methods measure distances between data points: a variable scaled to have higher variance than other variables will dominate such analysis. Pre-treatment states what levels of variance are significant in each variable, and which data points are fit for the analysis. This requires inclusion of process understanding in the data analysis process. This paper proposes the inclusion of this knowledge by using the logarithmic sigmoid (LogSig) function. The proposed method is compared to the common method of scaling variables to unit variance by studying data from a GSM-network using the self-organizing map (SOM).