Research And Implementation Of Uncertain XML Streams Classification Based On UOS-ELM

With the great development of the network,database and IT technology,in many applications,such as Internet management system,real-time sensor signal analysis system,telecommunication system and financial system,massive data are generated as high-speed dynamic XML stream.In the process of practical application,some reasons,such as unstable network,slow speed of information update,incomplete data collection,etc.,may cause uncertainties in the XML data stream.With the analysis on the uncertainty of the XML documents,this thesis proposes a baseline classification algorithm US-ELM over uncertain XML stream based on Extreme Learning Machine(ELM).We utilize the sliding window technique to process the stream data,and generate all the instances of each uncertain XML document in the stream.In US-ELM,the sliding window slides over one element at a time.In each sliding window,US-ELM uses the latest element to train a classifier using ELM.Then on the basis of US-ELM,this thesis puts forward an ensemble based ucertain stream classification algorithm EUS-ELM.In EUS-ELM,the sliding window slides over a unit of the size of the sliding window,and trains a number of classifiers based on ELM in the initialization phase with the initial data.We also put forward a concept of classifier uncertainty to detect whether the concept dift happens with the current data so that a new classifier needs to be trained.In the testing phase,this thesis introduces voting mechanism to deal with classification results.Finally,on the analysis of the EUS-ELM algorithm,this thesis introduces the incremental learning ideas to solve uncertain XML stream classification problem.We proposes an uncertain XML stream classification algorithm UOS-ELM based on OS-ELM.As can be seen from the experimental results,with the increasing size of the sliding window,the classification performance of EUS-ELM and UOS-ELM is better than UC-ELM.When the sliding window is small,US-ELM achieves better training time than UC-ELM and UOS-ELM;when the sliding window becomes larger,UOS-ELM and EUS-ELM have less training time than UC-ELM.As to EUS-ELM,along with the increment of the classifiers number,the classification performance of EUS-ELM gets better.Along with the increment of the threshold value,the classification performance of EUS-ELM gets better.Furthermore,UOS-ELM achieves better and better classification performance as the XML stream keeps flowing.