Research Of Multi-granulation Knowledge Acquisistion Methods Based On Rough Set

Since the 20 th century, we have gradually entered the Big Data Time. The trend of Big Data prompts data analysis and knowledge acquisition to become more urgent. Rough sets theory is a mathematical tool for analyzing and extracting knowledge from existed information without prior knowledge. The granular computating model based on rough sets theory, which simulates human thinking from different views of multi-granulation, can analyze and solve problems. The knowledge acquisition, including computing partitions of domains, attribute core, attribute reduction and rule acquisition etc, has always been a hot issue in rough set theory. Based on multi-granulation model of rough set, the related research works are presented as follows:1) Indiscernible relation is a key concept in rough set model and it can directly induce a partition of domain based on condition attributes. Firstly, an algorithm of quickly computing the partition of domain based on condition attributes is proposed in this thesis. All of condition attributes are assigned binary values, and these values are converted to an attribute layer by summing. Then the partition can be obtained by judging whether these sums are equality. Sencondly, an algorithm of quickly computing attribute core is proposed based on the above algorithm. It is realized through the complementary of the original information system. And it always remains high efficiency whether the information system is consistent. An example illustrates the detail steps of the proposed algorithms. Finally, experimental results show that the new algorithms are not only exact but also efficient.2) A heuristic algorithm for attribute reduction is proposed based on the above algorithms. It defines the reduction with approximation sets and knowledge granules. The discriminating power of the positive and negative regions is regarded as heuristic information. Then the condition attribute with the maximum discriminating power is added to the attribute core until the attribute reduction is obtained.3) An algorithm for mllti-granulation rule acquisition is proposed based on the above algorithm for computing the partition of the domain. The algorithm is running with simulating human thinking and acting ability with the changing granularities, and it can directly select rules with the methods of making marks in a position light. The number and average length of rules are as optimized as possible before ensuring the accuracy and coverage. Finally, an example and experiment results illustrate the steps of the proposed algorithm and its efficiency, respectively.