Heuristic Mode Research Of Decision Tree And Its Application In Attribute Reduction

Decision tree, generated in the 1960s, was raised by Hunt in the study of conceptual modeling. Decision tree, as a simple classification algorithm, is an effective tool for mining knowledge rules, it can deduce the classification rules in the form of decision tree from a set of examples, it is one of the most widely used logic methods. In recent years, it has been successfully applied in many fields, such as scientific experiments, medical diagnosis, weather forecasting, credit, audit, business forecasts, case detection and so on. Nowadays, many methods have already gotten some good classification results in some aspects. Although these methods are superior to the traditional entropy-based ID3 algorithm to a certain extent, there still exists further exploration space about the selection of heuristic function, the interpretability of the decision tree and the application. In this paper, we have done the following work.Firstly, based on the analysis of the essential characteristics of decision tree algorithm and the selection model of the expanded attributes for ID3, for the selection problem of expanded attributes, we explain the features that good expanded attributes should have by concrete graph, summarize the basic principles of expanded attributes selection, and put forward leaf criterion which can recognize the extension ability of attributes, data utilization criterion and comprehensive effect criterion, and establish a mathematical model with structural characteristics which can evaluate the extension capabilities of attributes.Secondly, by introducing the definition of quasi-linear function and discussing the properties of it. we give a selection model of expanded attributes based on quasi-linear function (denoted by QASM. for short), and discuss the performance of QASM from the theoretical and experimental aspects.Finally, based on the study of the existing attribute reduction algorithm, combining with the advantages of decision tree algorithm which is simple, fast classification and do not need to know a lot of background knowledge, we proposed the normal description pattern of rule knowledge. Based on the attribute reduction method of rough set. we give an attribute reduction method based on decision tree algorithm, then we analyze the basic characteristic and performance of the algorithm through an example, the results show that this algorithm is simple and strongly operable, and can effectively handle the attribute reduction problem of large-scale database.