基于二级聚类的家宽异常数据分析与应用

为充分挖掘家宽异常数据信息、直观了解家宽异常情况，针对用户端反馈家宽异常事件相关指标的具体数据，提出了一种基于异常主导因素和严重程度的综合评价体制，在此基础上分别利用余弦相关性和曼哈顿距离对归一化后的常见异常数据进行二级聚类，最终对常见家宽异常数据进行类别划分。现网数据实践表明，该方法可将大量无规则家宽异常数据划分为特点各异的若干类，并提取出需重点整改信息，并可建立一个预测准确率较高的KNN分类模型。

Analysis and application of home broadband anomaly data based on two-stage clustering

To fully explore abnormal home broadband data, perspicuously understanding abnormal home broadband situation, considering the specific user feedback data of indicators related to abnormal home broadband events, a kind of comprehensive evaluation system including dominant factors and the severity was proposed, and on this basis, with a hierarchical clustering using cosine correlation and Manhattan distance of normalized data respectively, the abnormal data was divided into several different categories. Practice with data from existing network shows that the proposed method can effectively divide a large amount of irregular abnormal home broadband data into different characteristics of several classes, extract the notable abnormal information, and can establish a KNN classification model with high prediction accuracy.