一种基于遗传算法的受限制的分类器学习算法

Constrained classifier learning algorithm based on genetic algorithm

提出了一种基于遗传算法的受限制BAN分类器算法-GBAN(genetic algorithm based BAN)。新算法采用了遗传算法进行网络结构的学习,限制了所学习的BAN分类器结构的复杂度。同时对TAN分类器的结构进行了扩展,得到了一种受限制的BAN分类器。针对这种分类器的结构学习,设计了结合对数似然的适应度函数及相应的遗传算子,并给出了网络结构的编码方案,使得该算法能够收敛到全局最优的结构。实验结果表明,当数据集属性之间关系相对复杂的时候,GBAN比TAN的分类准确率高,分类效果较好。

A restricted BAN classifier learning algorithm- GBAN based on genetic algorithm is proposed. Genetic algorithm was used to study the network structure. The structure of TAN classifier was extended by restricting the complexity of the structure of BAN classifier. And then a restricted BAN classifier is obtained. As far as this classifier＇s structure studying, the fitness function based on logarithm likelihood was designed. The code scheme of network structure , and the corresponding genetic operators are designed. As a result, the algorithm converges on the overall optimal structure. The experimental result indicated that GBAN algorithm has good classifying effect and is more accurate than TAN classifier when the relationship between attributes of a data set is relatively complicated.