Mining Incomplete Data with Many Missing Attribute Values A Comparison of Probabilistic and Rough Set Approaches
Patrick G. Clark, Jerzy W. Grzymala‐Busse, Martin Kuehnhausen · International Conference on Intelligent Systems · 2013
In this paper, we study probabilistic and rough set approaches to missing attribute values. Probabilistic ap- proaches are based on imputation, a missing attribute value is replaced either by the most probable known attribute value or by the most probable attribute value restricted to a concept. In this paper, in a rough set approach to missing attribute values we consider two interpretations of such value: lost and do not care. Additionally, we apply three definitions of approximations (singleton, subset and concept) and use an additional parameter called . Our main objective was to compare probabilistic and rough set approaches to missing attribute values for incomplete data sets with many missing attribute values. We conducted experiments on six incomplete data sets with as many missing attribute values as possible. In these data sets an additional incremental replacement of known values by missing attribute values resulted with the entire records filled with only missing attribute values. Rough set approaches were better for five data sets, for one data set probabilistic approach was more successful.