PREDICTIVE COMPLEX EVENT PROCESSING USING EVOLVING BAYESIAN NETWORK

Abstract

In view of the real-time change of data distribution and poor performance of fixed models in the streaming big data environment, we propose a novel structure learning method based on evolving Bayesian network. On the basis of Minimum Description Length and Max-Min Hill-Climbing algorithm, we improve the process of model construction and take Gaussian mixture model and EM algorithm for reasoning. When learning the model structure from the event flow, our method supports incremental calculation of scoring metric, and when the data and the model do not match, it can modify the model in time and support the dynamic updating of the network structure. The experimental results show that the proposed method is superior to the traditional methods in dealing with synthetic data and real data, which is more accurate and applicable to the prediction of complex events.