A Valued Tolerance Approach to Missing Attribute Values in Data Mining

One of the newest approaches to missing attribute values in data sets is based on a valued tolerance relation. The valued tolerance relation method of handling missing attribute values was not yet experimentally compared with other methods. The main objective of this paper was to compare the quality of two methods handling missing attribute values, one of them was the valued tolerance method, the other method was the MLEM2 approach, using the same interpretation of missing attribute values but a different approach to computing approximations and rule induction. Both methods were compared using not only an error rate, a result of ten-fold cross validation, but also complexity of induced rule sets. Our conclusion is that neither of these two methods is better in terms of the error rate. However, the MLEM2 approach produces, in most cases, less complex rule sets than the valued tolerance method.