先验知识引导下的基于遗传搜索的知识发现方法

PRIORI KNOWLEDGE-GUIDED KNOWLEDGE DISCOVERY METHOD BASED ON GENETIC SEARCH

针对知识库的不完备所导致的分类器泛化能力较差的问题,提出一种在先验知识引导下基于遗传算法的知识发现方法。该方法通过引入问题近似先验领域知识,进行种群初始化和变异函数构造,利用先验知识引导下的遗传算法对问题的解空间进行搜索,最终获取新知识。利用该方法可以获得同时覆盖先验领域知识和训练样例的一般知识,进而提高分类器的分类性能和泛化能力。实验结果表明,与经典遗传算法相比,不仅该算法的泛化能力更强,而且所获得特征规模较小。

Incomplete knowledge base leads to the problem of poor generalisation ability of the classifier. Aiming at this issue, we put for- ward a genetic algorithm knowledge-based discovery method guided by priori knowledge. The method initialise the population and construct the mutation function by introducing the approximate prior domain knowledge of the problem, makes use of the genetic algorithm guide by pri- ori knowledge to search the solution space of the problem, finally obtains new knowledge. Using this method, it is able to get the general knowledge covering the priori domain knowledge and training examples simultaneously, thereby improves the classification performance and generalisation ability of the classifier. Experimental results show that, compared with classical genetic algorithm, the algorithm has better gen- eralisation ability, and the features obtained is also smaller in size.