Information Metrics And Decision Trees Based On Neighborhood Equivalence Relation

The classic ID3 decision tree algorithm is suitable for discrete data classification,but it needs data discretization when it is used for continuous processing,which easily leads to information loss.Neighborhood rough sets can measure continuous data from the bottom layer without data discretization.The original data information can be largely preserved.Therefore,this thesis combines the neighborhood rough set theory and the decision tree algorithm,and comprehensively considers the advantages of the two to construct an algorithm for continuous data classification.The relevant content involves the following two aspects.1.The main content of this part is the neighborhood ID3 decision tree algorithm based on the neighborhood equivalence relationship.The shortcomings of the existing neighborhood relationship are analyzed,and then the neighborhood equivalence relationship is proposed.Based on the neighborhood equivalence granulation,the neighborhood information metric is constructed as Neighborhood information gain,and then construct the neighborhood ID3(NID3)decision tree algorithm,the NID3 algorithm improves the ID3 decision tree algorithm,which can directly implement continuous prediction and obtain better classification results.Both case analysis and data experiments show that the NID3 algorithm has the effectiveness of continuous data classification prediction,and is superior to the ID3 algorithm in classification machine learning.2.The main content of this part is the neighborhood decision tree of variable-precision neighborhood approximate equivalence granules.The variable-precision neighborhood approximate equivalence granules on the neighborhood decision information system are mined,and the related properties are discussed;To measure the uncertainty of the neighborhood decision information system,the neighborhood Gini index metric is used to measure the uncertainty of the neighborhood decision information system;finally,the neighborhood Gini index metric is used to induce the selection conditions of tree nodes,and the variable-precision neighborhood approximate equivalence particle is used as the tree splitting rule,so as to construct the NDT Algorithms.The results of experiments on the UCI data set show that the accuracy of the NDT algorithm is generally higher than that of the ID3 algorithm,CART algorithm,C4.5 and other algorithms,and it can be seen that the proposed NDT algorithm is effective.In summary,by combining the advantages of neighborhood rough set theory and decision tree algorithm,two new neighborhood-based decision tree algorithms NID3 and NDT are constructed,and they generalize and improve the classic decision tree algorithm.