Research On Field - Driven Human - Machine Interactive Decision Tree Model And Its Algorithm

Data mining is an important procedure of knowledge-discovery in database.Traditional data mining methods based on data object focuses on the efficiency of algorithms and accuracy rate of models, so it calledas a faced data or data drivendata mining of which results include lots of redundancyrules, common and normal knowledge even some mistaken conclusions obeyed common sense with lots of drawback.Decision tree is the main model for data mining classification techniques. Aiming at the problems of non-consideration mining users preferences, knowledge constrain of corresponding domain and operability of mining results, DDDM(Domain Driven Data Mining) thoughts puts forward to man-machine interactive decision tree model for construction decision tree process and results remark standard with the user interest and domain knowledge to remark the detail benefit in practice and deliver the decision support used in actual behavior.Man-machine interactive decision tree of domain driven insists on user participation degree and initiative in data mining to change the only using pre-set algorithm as hidden mode concentrated in machine automatic searching of traditional decision tree, the data mining system development of mode input becomes factual issues of decision making system for solving the solution output.The main researches of this thesis include the following parts:1.Based on traditional data mining confined to the data and algorithm without constraint from varied domain knowledge in real environment, the inadequacy and drawback of old mining mode are analyzed to summarize the development process of domain driven data mining for decades and the researches results of professors at home and abroad from the aspects of theory level, methods framework and mode structure.2. Cost matrix of design transforming attribute value and efficiency matrix of classificatory display the user interestingness and mined domain knowledge restraint. The concept and value of decision branches use as the feature selection measurement for domain driven decision tree algorithm.3.According to classical decision tree algorithm as basic framework, a kind of man-machine interactive decision tree algorithm and decision recommended algorithm of evaluationtransforming effective branches are put forward to make user obtain the actual benefit of the best solution and to make accurate, reasonable and operational decision.4.Based on the above-mentioned researches, T-Systems broadband customer data of Deutsche Telekom AG makes a demonstration. According to the user interestingness and complete business profit, cost matrix of design transforming attribute value and efficiency matrix of classificatory mines same test data to analyze and compare the actual business benefit of mining results. The experiment shows man-machine interactive decision tree of domain driven without decreasing classification accuracy condition, the mining decision rules is easy to understand and have strong practice, high business value.