决策树英语

Classification of knowledge acquisition are data mining to achieve one of the important tasks, the core problem to solve classification model are constructed and classification algorithm implementation. In this paper, decision tree classification method has the representation of the method C 4.5 as an example, the introduction of data mining in a classification of one decision tree classification method and its building and applied research. Keywords: C4.5 decision tree data mining 1 Introduction Generated from the data classifier is an effective way to set up the decision tree (Decision Tree). Express decision tree method is the most widely used logic one way, it was no order from the one without the rules of reasoning examples express the form of decision tree classification rules. Decision tree classification method used to top-down recursive way, in the decision tree internal node property value for comparison, and according to different property values to determine the next node from the branch, in the decision tree leaf nodes conclusion. From the decision tree root to the leaf nodes on a path corresponds to a Conjunctive rules, decision tree on one group corresponds to the rules of disjunctive expression. 2 the basic concepts of decision tree Decision tree classification is divided into two types of trees and regression tree, classification tree for discrete variables make the decision tree, regression tree for continuous variables make the decision tree. General data mining tool, select split permit conditions and pruning rules, as well as control parameters (the minimum node size, the greatest depth of the tree and so on) to limit the decision tree. Decision tree as a tree, the tree root node is the entire data set space, each node is a single variable test, the test data set space is separated into two or more blocks. Each leaf node are classified under a single record. Decision tree for the purpose of producing a collection of training set for each record must have a good sub-category. Property to determine which domain (Field) as the classification of the best indicators. General practice is exhausted all the property domain, for each domain to split the property to quantify the quality of calculated one of the best split. Different algorithm for calculating the standard split-domain property is not the same. Second, repeat the first step, until every leaf node within the records belong to the same category, up to a complete tree. The purpose of the decision tree structure is to identify the property and type of relationship question, use it to predict the future unknown types of record types, this function has a prediction system called decision tree classifier. At classification analysis, the decision tree model is the most popular model, mainly because it is very easy to use graphical (tree structure) approach the performance of excavation results, adapted to enterprise management to make a decision. 1986 J. Machine LearningJournal at Ross Quinlan delivered a speech entitled "Indudion of Decision Trees" thesis, the introduction of a new ID3 algorithm. On this basis, he also carried out on the ID3 algorithm to add and improve, put forward a more advanced algorithm C4.5. This article discusses how to apply the C4.5 decision tree algorithm for structural classification of customers. 3 decision tree generated by the process of Decision tree is a flow chart similar to several structures, one of each internal node, said at a property on the test, each branch represents a test output, and each leaves the representative of the class or category of point distribution. (1) To collect customer information, information of data to merge, forming the structure of a unified customer information data sources. (2) of the data source for data pre-processing, removing the decision-making has nothing to do with the property and the high branches of property, value-type property will be generalized to deal with vacancies, as well as the value of property to form a decision tree training set. (3) of the training set for training, computing the information gain of each property and access rate, access to the rate of selection but at the same time The largest information gain access to the property not less than the average of all the property, as the current master node property for the The property of every possible value to build a branch. The sub-node contains a subset of samples to implement the above recursive process, until the subset of data records in the main on the property values are the same, or not for the division of property can be used to generate the initial decision tree. (4) of the initial decision tree to tree pruning. Mainly after the pruning algorithm to generate the initial decision tree for pruning, and pruning using a pessimistic estimate to compensate for the tree to generate optimism at the time of deviation. (5) obtained by the decision tree classification rules extraction. From the root to the leaves of each path to create a rule to form a rule set. Rule set will be displayed to the user, the user filter rules have been credited to consider possible rules database. (6) When a new customer at the company register, the system derived from the use of decision tree rules for the new client information for data analysis, prediction of the behavior of customers belong to which class, and thus for the company's customer marketing strategy to support decision-making.