High Efficient Algorithm For Knowledge Reduction Based On Granular Computing And Method For Dealing With Missing Value

It always exists noisy value or missing value in original data, these problems impact results of the data mining and machine learning. There are exist many methods which are based on statistics principle, but these methods are fit for estimating mode parameters not for filling missing values, they are mainly used for unsupervised data set. At present, there are a little methods for decision table and these methods may generate rules with lower support and lower accuracy. In this paper, a method for dealing with missing value based on minimum description length principle is proposed, a case study has been performed, proving it can generate rules with higher support and higher accuracy. Research on high-efficient algorithm for knowledge reduction is another important content in this paper. It is very important means for efficient rules extracting. At present, almost all algorithms for knowledge reduction are based on Rough sets theory. Rough sets theory has Irreplaceable advantages in dealing with uncertainty knowledge, and knowledge reduction is one of the most important contents in RS theory. The high efficient reduction algorithm is very important to large-scale data sets, but many reduction algorithms are suitable for the small-scale data. This paper gives a high efficient algorithm for knowledge reduction based on granular computing. The analysis indicates the time complexity is small and the running time reduces greatly, moreover this algorithm can produce minimum reduction in most situations.