Bayesian classification algorithm of dynamic data stream based on bootstrap
XU Chonghuan · Computer Engineering and Applications Journal · 2011
Dynamic data streams have features of large data,instant change,costly random access and difficult storage of detailed data,so mining of such dynamic data streams puts forwards high requirements on the computing power and storage capacity.According to the above features,a Bayesian classification algorithm of dynamic data stream based on bootstrap is proposed to process and analyze dynamic data streams with the sliding window model.This model,taking data of each window as the basic unit,processes and analyzes the data of windows.The algorithm adopts the bootstrap method to cut and optimize the attributes of data to be classified,solving the problem in multi-linear inter-relation between data attributes.The algorithm,combining characteristics of Bayesian algorithm,adopts the dynamic incremental storage tree to store the dynamic sample data stream to realize the static finite storage of infinite dynamic data streams without distortion of information and ultimately solve the biggest problem in dynamic data stream mining——data storage.The all-Bayesian classifier and k-Bayesian classifier are adopted to classify the optimized data,and their updates are made according to the features of data streams. This algorithm overcomes the attribute independence of the Bayesian classifier and its limitation only to the static data.It overcomes the biggest problem of dynamic data stream——the data storage.Experimental tests prove that the Bayesian classification algorithm based on bootstrap has high timeliness and accuracy.