Research On The Method Of Decision Rule Acquisition Based On Concept Lattice

As a kernel data structure in the theory of formal concept analysis, concept lattice is a powerful tool for data analysis, used to extract hidden knowledge pattern in data. There are many advantages to build and apply concept hierarchy in the knowledge acquisition, meanwhile, Hasse diagram of concept lattice embodies such concept hierarchy structure and reflects the generalization and specialization between the concepts. Therefore, studying the basic theory of concept lattice and applying it in knowledge discovery have important significance.This thesis mainly analyses and compares several construct algorithms for concept lattice, and study decision rule acquisition based on concept lattice, meanwhile some significant results are abtained.In the aspect of analysis and comparison on several construct algorithms for concept lattice, in this thesis, several typical constructing algorithms are summarized and discussed. Among them, incremental algorithm is stable and efficient when the formal context contains large objects, on the contrary, batch algorithm is not stable enough, so it is suitable to use when the formal context contains little objects.In the aspect of rule acquision based on concept lattice, a new algorithm for mining decision rules is proposed based on the completeness of all nodes in concept lattice, decision rules are computed for given support degree and confidence degree threshold. Based on the above algorithm, an optimized algorithm is given. In this algorithm, closed label of the node which includes decision attribute is discussed,then decision rules with short antecedent and same support degree as the above algorithm are computed. We implement the algorithms above and select a decision table from nursery database of UCI database and the results illustrate that the algorithm is valid.In this thesis, several typical constructing algorithms are analyzed and compared, and the result has important significance to study new construct algorithm. On the other hand, a new algorithm for mining decision rules based on concept lattice could compute reasonable decision rule set when different threshold is given. In addition, concept lattice could be updated conveniently when objects are added or deleted in the decision table, it's not necessary to construct the lattice again, so the algorithm is flexible.