Approach of machine-learning-based visualization for the evaluation of fuzzy effects of low-dose radiation

Kanzaki, Norie   ; Sakoda, Akihiro   ; Kataoka, Takahiro*; Yamaoka, Kiyonori*

It is not easy to evaluate fuzzy effects of low-dose radiation by basic statistical analysis or basic machine learning. In the present study, the modification of self-organizing maps, which is a kind of machine learning, was made for the evaluation of such effects: namely many reference vectors which learned input dataset by self-organizing maps were reanalyzed with the same technique. Based on this procedure, we analyzed a dataset about low-dose radiation as well as a benchmark dataset, suggesting that the modified self-organizing maps can work even for input data with complex topology and data distribution.