A novel fuzzy oblique decision tree based on AFS theory and fuzzy information entropy
Based on the Axiomatic Fuzzy Set(AFS)theory and fuzzy information entropy,a new Fuzzy Oblique Decision Tree(FODT)algorithm is proposed in this paper,and it mainly includes four aspects.Firstly,the AFS theory is used to determine fuzzy membership functions automatically according to the raw data distribution.Secondly,we apply the fuzzy information entropy to guide the polymerization process of the fuzzy concepts,which not only takes into account the similarity between the polymerization results and the target category,but considers the dissimilarity between the polymerization results and other categories.Thirdly,the FODT considers a fuzzy rule with multiple features at each non-leaf node while the standard axis-parallel decision trees take a single attribute into consideration.Finally,the threshold is optimized by genetic algorithm to control the balance between the tree size and classification accuracy.The construction of the FODT consists of four major steps:(a)generate membership functions by AFS theory,(b)extract fuzzy rules by the fuzzy rule extraction algorithm(FREA),(c)construct the FODT by the fuzzy rules obtained from(b),and(d)determine the optimal threshold resulting in a final tree.Compared with five traditional decision trees(C4.5,LADtree(LAD),BFTree(BFT),SimpleCart(SC),and NBTree(NBT))on eight UCI machine learning data sets,the experimental results demonstrate that the proposed algorithm outperforms the traditional decision trees in terms of both classification accuracy and the size of the trees.
AFS Theory Fuzzy Information Entropy Oblique Decision Tree Genetic Algorithm
CAI Yuliang ZHANG Huaguang HE Qiang SUN Jiayue
College of Information Science and Engineering,Northeastern University,Shenyang,110819,P.R.China College of Computer Science and Engineering,Northeastern University,Shenyang,110819,P.R.China
国内会议
厦门
英文
122-128
2017-11-17（万方平台首次上网日期，不代表论文的发表时间）