Research On Parallel Bayesian Network Classifier Based On C-MCMC And MapReduce

Bias network classifier has strong learning and reasoning ability,and it is one of the research focuses in the field of data processing.Although the Bayesian network classifier classification showed good prediction performance,but there are still only a priori the utilization rate is not high,and practical difference in learning can not be optimal network structure,thus affecting the performance of the classifier.How to better implement the parallelization of the existing Bayesian network classifier is still one of the most urgent problems to be solved.In order to solve the above problems,a parallel Bayesian network classifier is designed and implemented in this paper,including the following two parts:(1)Based on the Markov Chain Monte Carlo algorithm(Markov Chain Monte Carlo,MCMC)on the basis of the introduction of prior knowledge,Bias proposed a new network structure learning algorithm of C-MCMC(Constrained-MCMC),and the effect of using prior knowledge of algorithm MCMC Bias network structure learning,and the effectiveness of the algorithm is verified through a series of the experiment,to learn from Bias network more excellent.(2)The C-MCMC Bias network structure learning algorithm is applied in the traditional augmented Naive Bayesian classifier(BAN)and general Bias(GBN)network classifier,and estimate the corresponding parameters,so as to design the C-MCMC BAN classifier and C-MCMC GBN classifier.With the parallel programming model MapReduce of open source platform Hadoop,the corresponding Map function and Reduce function are designed.The C-MCMC Bayesian network classifier using MapReduce parallel programming frameworkfor parallel programming,gives the specific implementation process,and through the construction of Hadoop platform to verify the improvement of the efficiency of the parallel algorithm and improved algorithm.The experimental results show that the Bayesian network classifier is designed in this paper is superior to the traditional,has a higher classification accuracy and efficiency,and is suitable for large data processing applications,can be used in many occasions,it has broad market prospect.