Holistic data accuracy assessment using search & scored-based Bayesian network learning algorithms

Data accuracy is one of the central criteria for data quality. Bayesian network (BN) learning algorithm based on independence test is a promising approach for assessing data accuracy. However, it may get stuck in out-of-memory exception when confront with large network and small percentage of inaccurate data. In this paper, we extend the method for data accuracy assessment to search & scored-based BN learning algorithms. This category of algorithms can overcome the shortage of the former one, which will broaden the application of BN learning algorithms in assessing data accuracy. More indicators based on adjacent matrix are developed to view data accuracy from different perspectives. Two algorithms are tested to verify the efficacy of our research and the results show that the mainstream search & scored-based BN learning algorithms scale well with the strategy for data accuracy assessment.