Cluster Analysis And Its Application Study Based On Unsupervised Featrue Selection

With the widespread use of mobile Internet,social media and other emerging technologies,amounts of media data has been generated,which contains rich social information and economic value.Media data is variety,timeliness,short update-interval and low value density.These features greatly increase the data processing needs for text,image,video,audio and others in various fields.Therefore,accurate and efficient processing for media data plays an important role in academic research and economic prediction.This paper focuses on social data on the internet and mobile communication data,studies unsupervised feature selection cluster and its application for dynamic stream data.The major work is as follows:1.An unsupervised feature selection algorithm for dynamic network media based on user correlation(UFSDUC).Firstly we classified the relationship of users in social networks as the constraint of unsupervised feature selection.Then the Laplacian operator is used to build the feature selection model of user relevance,quantify the strength of relationship between users.The optimal user relationship in the feature model is given by Lagrangian multiplier method.Finally,the threshold of streaming media data is set according to gradient descent method,for calculating the non-zero feature weights to update the optimal subset of features,furthermore to effectively classify data.The proposed algorithm can accurately select features in dynamic media-data network in real time on the foundation of keeping user correlation integrity.2.A method for traffic anomaly detection with incomplete data(TAD).Firstly,analyze the cluster according to the correlation between real-time mobile phone data and vehicle density,which can improve the computational efficiency of incomplete data.Secondly,detect distributed dynamic events by using the changing rate of mobile phone call,capture road abnormalities in time.Finally,through tracking the development of abnormal events,obtain abnormal distribution routes,then estimate the influenced area and take effective regulatory measures.The experimental results show that the proposed method can effectively improve the detection efficiency of traffic anomaly events,has innovativeness.Media data transmission is developing toward supersizing,dynamic and high-dimensional,feature selection and clustering analysis can obtain the value information,improve the accuracy and speed of feature extraction and clustering calculation.The two research programs in this paper propose innovative solutions to the research of media data,which can alleviate the contradiction between speed and accuracy.