Based On Markov Blanket Bayesian Network Discovery Algorithm Study

Markov blanket discovery plays an important role in both Bayesian networkinduction and feature selection for classification tasks. Markov blanket discoveryalgorithm can learn and discovery the Markov blanket of target variable from the actualdata sets. Markov blanket discovery algorithm can be divided into two categories.However, their principles are not the same. The first categorys of algorithms as therepresentative to IAMB are very fast, but of low accuracy. The second categorys ofalgorithms as the representative to IPC-MB are very accurate, but slowly. Therefore,both of the algorithms are of practical value.Bayesian network is a type of statistical models that efficiently represent the jointprobability distribution of variables. The paper will research and design Markov blanketdiscovery algorithm based on Bayesian network. The main work is as follows:(1) Propose a novel Markov blanket discovery algorithm, called DynamicOrdering-based Search (DOS) algorithm. DOS algorithm bases on the IPC-MBalgorithm and compensates for the shortcomings and deficiencies of the IPC-MB.Through the integrated use of “sort”,“filter”,“optimize the symmetry principle” ofthree innovative trategies, DOS algorithm can greatly reduce the number of conditionalindependence tests and improve the speed and accuracy of discovering the Markovblanket. Firstly, this paper details the principle and implementation process of DOSalgorithm for discovering Markov blanket and analyzes its main features. Secondly, thispaper theoretically proves DOS algorithm on its correctness and analyzes timecomplexity of DOS and IPC-MB algorithm in form of contrast. Finally, large, repeatedexperiments show the excellent performance of DOS algorithm.(2) Propose a novel Markov blanket discovery algorithm, called Improve-IAMBalgorithm. Improve-IAMB algorithm that bases on the algorithm IAMB is characterizedby the strategy of the “group”. The strategy reduces the number of conditionalindependence tests and inhibits the negative impact of data sets noise. Improve-IAMBalgorithm improves the speed and accuracy of discovering Markov blanket. Firstly,this paper details the principle of Improve-IAMB algorithm for discovering Markov blanket and describes its implementation process of with an example. Secondly, thispaper analyzes time complexity of Improve-IAMB and IAMB algorithm in form ofcontrast. Finally, empirical results show that the Improve-IAMB algorithm performsmuch faster and more reliably than the IAMB.