New Algorithms of Feature Selection and Big Data Assignment for CBR System Integrated by Bayesian Network
Yuan Guo, Yu Xin Sun, Kai Wu, Kerong Jiang · ACM Transactions on Knowledge Discovery from Data · 2020
Under big data, the integrated system of case-based reasoning and Bayesian network has exhibited great advantage in implementing the intelligence of engineering application in many domains. To further improve the performance of the hybrid system, this article proposes Probability Change Measurement of Solution Parameters (PCMSP)–Half-Division-Cross (HDC) method, which includes two algorithms, namely PCMSP and HDC algorithm. PCMSP algorithm can select principal problem features according to their effects upon all solution features measured by calculating the weighted relative probability (RP) change of all solution features caused by each problem feature. PCMSP algorithm can perfectly work under big data no matter how complex the data types are and how huge the data size is. HDC algorithm is used to assign the computation task of big data to enhance the efficiency of the integrated system. HDC algorithm assigns big data by grouping all the problem parameters into many small sub-groups and then distributing the data which covers the same sub-group of problem parameters to a slave node. HDC algorithm can guarantee enough efficiency of the integrated system under big data no matter how large the number of problem parameters is. Finally, lots of experiments are executed to validate the proposed method.