Affiliations: Department of Electrical Engineering and Computer Science, University of Kansas, Lawrence, KS 66045, USA. [email protected] | Department of Electrical Engineering and Computer Science, University of Kansas, Lawrence, KS 66045, USA and Institute of Computer Science, Polish Academy of Sciences, 01-237 Warsaw, Poland. [email protected]
Note: [] Address for correspondence: Department of Electrical Engineering and Computer Science, University of Kansas, 3014 Eaton Hall, 1520 W. 15th St., #2001, Lawrence, KS 66045-7621, USA
Abstract: In this paper we present results of experiments on 166 incomplete data sets using three probabilistic approximations: lower, middle, and upper. Two interpretations of missing attribute values were used: lost and “do not care” conditions. Our main objective was to select the best combination of an approximation and a missing attribute interpretation. We conclude that the best approach depends on the data set. The additional objective of our research was to study the average number of distinct probabilities associated with characteristic sets for all concepts of the data set. This number is much larger for data sets with “do not care” conditions than with data sets with lost values. Therefore, for data sets with “do not care” conditions the number of probabilistic approximations is also larger.
Keywords: characteristic sets, singleton, subset and concept approximations, lower, middle and upper approximations, incomplete data
