遗传算法中基于规则的分类器编码长度研究

Research on Classifier Encoding Length Based on Rule in Genetic Algorithm

遗传学为基础的机器学习使用遗传算法作为学习机制,设计以规则为基础的分类系统,通过训练数据集来实现类别的精确描述。针对遗传算法编码没有统一标准的问题,研究基于规则的分类器个体特征编码长度与分类准确率以及效率之间的关系,通过概率逼近分析个体特征编码长度对分类准确率的影响,利用迭代步骤数的数学期望计算方法,计算遗传算法分类器的分类效率。实验结果证明,遗传算法在密西根编码条件下,个体特征编码长度越长,分类器的分类准确率越高、收敛速度越慢。

Genetic Algorithm（GA） is used as a machine learning tool for designing linguistic rule based on classification systems, accurate description of the category is category is generated by the training data set. So far there is no a uniform standard for the problem of the encoding of GA, this paper researches on the relationship between the individual characteristics coding length, the classification accuracy and the efficiency of classifier. It analyzes the effect of the coding length for classifier classification by probabilistic approximation, uses the method of getting the number of iteration steps mathematical expectation which is used to calculate the GA of classification eff＇lciency. Experimental result shows that the longer encoding length is, the higher accuracy and slower convergence rate are for GA under the condition of Michigan coding.