Study Of Attribute Reduction Algorithms Based On Binary Discernibility Matrix And Of Rule Extraction

Facing the constantly increasing data, it is important and significant for people to find an efficient tool, which can find some interesting and useful knowledge from tremendous information databases. Then we can make decision in aspects of living and producing basing on them.Rough Set Theory is proposed firstly by Z.Pawlak in 1982. It is an important tool for effectively analyzing various kinds of the incomplete and imprecise information, which can analyze and reason the data, discover the connotative data and find out the latent rules. Finding the reductions and the core of Information System plays an important role in Rough Sets. Attribute reduction is one of the most important steps in it. It is a process of crossing out the dispensable attributes in the information system, while keeping the capability of classifying. This provides possibility for people to extract rules and provides people with supports of making decision. My project studies three main aspects which are shown as follows:Firstly, this paper introduces an attribute reduction algorithm based on binary discernibility matrix. We point out and improve two drawbacks in one of the existent algorithms. On one hand, since the core of rough set is an important basis in the Rough Set Theory in attribute reducing, we can predigest the initial discernibility matrix fast using the core attribute. While the core is not empty , especially when the core contains important attribute, this amelioration can reduce so much unneeded calculation that it can cut down the size of matrix in a large range. Therefore, it can improve the efficiency of algorithm effectively. On the other hand, we give a new method for calculating the attribute significance. This method considers the heuristic information which contains in the decision system and puts the most important attribute with the maximal weight. This assures that we select the most important attribute every time and get the optimal reduction of the decision system fastly.Secondly, we study the rule extracting and rule simplifying methods ulteriorly. We add a step in the rule extraction algorithm in order to avoid creating the contain rules and assure that we can extract the minimal rule set. At the last of this chapter, we introduce a method of the rule simplifying and find an error in it. So we study it carefully and point out the reason producing the error. Then, we present a new algorithm to simplify rules. This new method extracts rule which has the least condition or contains the most core-value in the rule core-set. We can gain shortest and uniteable rules based on the former selecting policy.Finally, we investigate into the ant colony algorithms and design an efficient global search and improve a new choice rule, which makes better pheromone based on the value of a target function. This algorithm can accelerate the search in the solution space and then improve the speeding of calculating the best solution.