Analysis on the fuzzy filter in fuzzy decision trees

Fuzzy decision tree, which is an extension of classical decision tree, is an effective method to extract knowledge in uncertain classification problems. The essential characteristic of fuzzy decision tree is to take of the value in [0,1] to describe the subordinate relation between examples and linguistic terms, This avoids the unreasonability generated by the classical decision tree, which only use 0 and 1 to describe the subordinate relation. By well using the relation between examples and attributes, the fuzzy decision tree holds high performance, but at the same time, this fuzziness leads to the uncertainty of classification and disturb the building of a good decision tree. So, how to filter fuzzy data properly is an important segment in the building and the application of decision trees. This paper analyzes the effect of fuzzy filter on the fuzzy decision tree by using many experiments, which are based on both fuzzy ID3 heuristic method and the minimal ambiguity based heuristic method. This paper also discusses the fuzzy filter level from both the inherent mechanism of building a fuzzy decision tree and the essence of fuzzy information filter, and proposes a practical method to determine this filter level for different data sets.