A Method Of Machine Learning And The Design Of Inference Engine Based On Decision Tree

The machine learning method for decision tree,it is an effective method to simulate mankind study behavior. The core issue is how to construct the topological structure of trees on the basis of known training sample data.Because machine learning problem is an issue which is closely related to problem domain, So different application target may lead to some topics can not be solved or lead to some topics more specific or deeper can not be well solved by existing theoretical frameworks. The work will develop a research work for machine learning methods of task-oriented nature that is taking auto transmission fault diagnosis problem as a background and ID3algorithm as the basic model.This paper firstly discussed a kind of machine learning algorithm models based on an ID3algorithm, its essence is to approach a group of discrete objective function. The function obtained after studying is represented as a decision tree, to search the possible decision-making space greedily with the top-down method.On this basis, aiming at the problem of automotive gearbox fault diagnosis, constructed a machine learning algorithm which taking information entropy as node splitting rule and taking node data quantity, information entropy contribution degree, structural depth as splitting and stopping criterion.On the basis of fundamental learning model,this paper further discusses pruning and the problem of conflict resolution for decision tree. Design of decision tree pruning criterions based on pessimistic error pruning and minimum error pruning are given,and show a new strategy design for conflict resolution.Based on the above work, this paper finally discussed problems of inference engine design of automotive gearbox fault diagnosis, which includes feature extraction, fault diagnosis knowledge bases and establishment of the rule bases, attribute tables, creation of decision tables, problems of the building of decision tree learning model. At the same time, a test and analysis was conducted to the established decision tree module.