Many real-life data sets are incomplete, or in different words, are affected by missing attribute values.Three interpretations of missing attribute values are discussed in the paper., lost values (erased values), attribute-concept values(such a value may be replaced by any value from the attribute domain restricted to the concept), and"do not care" conditions (a missing attribute value may be replaced by any value from the attribute domain). For in- complete data sets three definitions of lower and upper approximations are discussed. Experiments were conducted on six typical data sets with missing attribute values, using three different interpretations of missing attribute values and the same definition of concept lower and upper approximations. The conclusion is that the best approach to miss- ing attribute values is the lost value type.
展开▼
