英语论文网

本文针对汽车质量评价分类的实际需求，重点研究了决策树的核心技术：决策树建树算法和决策树剪枝算法。在建树算法上重点介绍了ID3、C4.5等6种典型算法的建树过程以及性能分析等，最后总结和分析全面比较每种建树算法的优缺点。在决策树剪枝算法上重点介绍了CCP、REP等5种剪枝算法的剪枝策略，通过比较总结了每种剪枝策略的特点。

In this paper, automotive quality assessment and classification of the actual needs , focusing on the core technology of the tree : tree pruning algorithm achievements algorithms and decision trees . Focus on the achievements algorithm introduced ID3, C4.5 and other six kinds of typical algorithms achievements and performance analysis, process , summarize and analyze the last full comparison of the advantages and disadvantages of each algorithm achievements . Focus on the decision tree pruning algorithm introduced CCP, REP pruning algorithms such as five kinds of pruning strategy , summarized by comparing the characteristics of each pruning strategy .

本文根据前人研究的成果分析出其中的不足，提出了“基于规则信息量和前件化简阈值的规则后剪枝算法”，该算法以REP剪枝策略为指导，使用规则信息量和可信度从原始规则中提取可信度高的有用规则，使用属性重要性和前件化简阈值对规则的前件进行简化。

Based on the results of previous studies which analyzed the deficiencies , proposed a " rule-based and informative antecedent simplification threshold rule pruning algorithm," The algorithm REP pruning strategy as a guide, the amount of information and use rules reliability rules extracted from the original high credibility of useful rules , the use of attribute importance and simplification thresholds antecedent antecedent of the rules can be simplified . REP algorithm through a comparative analysis of the proposed algorithm has less extraction rules , regulations and high reliability , high classification accuracy , the algorithm is high stability, flexible control , etc., to overcome the noise data and overfitting problems .

最后本文利用C#编程实现了决策树分类器，通过4组实验得到以下结论：（1）C4.5算法构建的决策树规模较ID3小，但受数据影响分类精度略低；（2）规则后剪枝比REP剪枝得到更少的优化规则，同时分类精度很高；（3）规则置信度在[0.3，0.7]区间、前件化简阈值在[0.8，1.4]区间时，提取的规则性能最佳；（4）优化后的规则能满足汽车质量评价分类的需求。

Finally, we use C # programming decision tree classifier, through 4 set of experiments the following conclusions : (1) C4.5 decision tree algorithm to construct smaller scale than ID3 , but affected by the data classification accuracy is slightly lower ; ( 2 ) rules pruning pruning get less than REP optimization rules , while classification accuracy is high ; ( 3 ) rule confidence in [ 0.3,0.7 ] interval , the former member simplification threshold [ 0.8,1.4 ] interval , the extraction rules best performance ; ( 4 ) optimized to meet the rules of automotive quality assessment and classification requirements.

Classification in data mining is a very important task , the current classification and prediction techniques are widely studied in data mining topics , and has been in the telecommunications , banking , insurance, retail, healthcare and many other industries are widely used, for the future commercial development and people's production and life have a profound impact . The purpose of classification is to learn a classification function or classification model, which is able to data in the database is mapped to a given item in a certain category . After the classification model can be predicted using historical data record automatically deduced for the promotion description given data to predict future data .

In data mining classification and prediction , there are decision trees, Bayesian classifier , back propagation classification, based on concepts from association rule minin