Fuzzy Decision Tree Induction Algorithm Based On Rough Set Technology

Decision tree is an effective method to solve classification problems, which has been successfully applied in pattern recognition, decision support, biological information processing and many other fields. ID3 algorithm is a well-known decision tree induction algorithm. The objects dealt with by ID3 are discrete-valued decision tables, but it cannot directly deal with the discrete-valued decision tables with fuzzy decision attributes. In order to solve this problem, based on the rough fuzzy set technology, we propose a fuzzy decision tree induction algorithm Rough Fuzzy Decision Tree(RFDT). RFDT uses rough fuzzy dependence as a heuristic to select the extended attributes, and uses the fuzzy entropy as the termination condition of the leaf nodes to construct the fuzzy decision tree. RFDT can deal with the problem mentioned above.The fuzzy ID3 algorithm is an extension of the ID3 algorithm. It is tailored for inducing fuzzy decision trees from the fuzzy decision tables with fuzzy condition attributes and fuzzy decision attribute. When fuzzy ID3 algorithm is applied to fuzzy decision tables with continuous-valued conditional attributes, it is necessary for fuzzy ID3 to fuzzify the continuous-valued attributes, but the fuzzification will result in losing useful information. In order to deal with this problem, based on the tolerance rough fuzzy set technique, an induction algorithm of fuzzy decision trees named Tolerance Rough Fuzzy Decision Tree(TRFDT) is proposed. TRFDT uses tolerance rough fuzzy dependence as a heuristic to select the extended attributes, uses fuzzy entropy to select the best cut point to construct the fuzzy decision tree. The advantage of TRFDT is that it can deal with the fuzzy decision table of continuous value directly, and the fuzzify process is not required.We experimentally compared the two proposed algorithms with fuzzy ID3 and FDTs respectively on 11 UCI data sets, the experimental results show that the two algorithms proposed in this thesis are effective and efficient.