一种提升树算法在网络故障关联分析中的应用

The Application of Boosting Tree Algorithm in Correlation Analysis on Network Fault

针对移动网络故障反馈与故障分类的准确率问题,提出一种基于k-NN的多分类器提升树算法,该算法采用k近邻对特征向量空间划分,结合adboost误差函数和迭代算法,实现多分类器决策判别,并利用R语言可视化方法的关系数矩阵提取和筛选出投诉/故障映射主属性,进而建立基于基站退服、覆盖盲点、非网络原因3类故障投诉关联分析及预测模型。实验结果表明,与C50决策树、RIPPER分类规则及SVM比较,在网络故障关联预测方面新算法的分类准确率提升约2%~10%。

According to the accuracy problems of mobile networks fault feedback and fault classification,multiple classifier boosting tree algorithm based on k-NN is put forward.The algorithm uses k neighboring to feature vector spacial classifcation and combines adboost error function and iterative algorithm to realize classifier decision discrimination,the model adopts the relation matrix of R language visualization method to extract and select the main attributes of complaint/fault map.Furthermore,the model uses the boosting method to realize the decision of multiple classifiers,and establishes three kinds including the base station retirement service,coverage blind spot and nonnetwork cause.The model fully expresses the cause of the user’s complaints and realizes the effective analysis and prediction of the relationship between operation complaints and the cause of the failure.The experimental results show that the classification accuracy of the model is about 2%~10%higher than that of C50 decision tree,RIPPER and SVM.