Formal Study On Key Issues In Classification Rule Mining Based On Formal Concept Analysis

By analyzing existing problems in the present methods of classification rule mining based on FCA, the paper points out that mining classification rules with incremental construction of lattice accords more with the characteristic of incremental learning in training stage. Equotient lattice is first introduced into the incremental construction of lattice and two incremental algorithms are designed based on equotient lattice which turn out to be more efficient comparing to Godin's algorithm and Bordat's algorithm in most situations. The design principle for pre-pruning concept lattice based on equotient semi-lattice is proposed. By using the the principle of synchronization computation of intent reduction advanced in the paper, the new algorithm based on equotient semi-lattice shows a well behavior and can be used to mine two type of classification association rules quickly. After introducing intensity constraint and support constraint, a new algorithm for pruning lattice is proposed and then conceptual rules are selected by use of another parameter:noise constraint, leading to a great decrease in the number of rules mined. A classification algorithm designed on the basis of conceptual rules mined has more accurate precision and less time used in rule mining than that of some classific methods, which confirms the idea of mining classification rules based on incremental construction of concept lattice proposed in the paper.