Abstract
The study is concerned with a development of information granules and their use in data analysis. The design of information granules exploits a concept of "overshadowing" meaning that, we retain a given level of membership to a given concept unless faced with a contrary evidence. A detailed algorithm is provided and illustrated through a number of numerical studies. The idea of noninvasive data analysis is then introduced and discussed from a standpoint of a minimal level of structural dependencies to be used in the model.