摘要： 针对尿沉渣红白细胞的特征选择问题，提出结合双向法的改进遗传算法，利用特征位逐步锁定法，结合小生境技术和自适应交叉变异算子共同缩小遗传算法的搜索空间。为了提高特征集的优选效果和稳定性，引入“多票投选”机制进行综合判断输出所求的最佳特征子集。实验结果表明，该算法优选的特征集与未进行特征选择和经过简单遗传算法(SGA)特征选择得到的特征集相比，识别率较高、特征数较少，反向传播神经网络(BPNN)分类器的维数复杂度明显减少。

关键词: 双向法, 特征位锁定, 特征选择, 遗传算法, 反向传播神经网络

Abstract: Based on red cell and white cell binary class space feature selection issue, this paper proposes one modified genetic algorithm for it. The modified genetic algorithm uses feature-fixing technology, and puts the Niche technology and adaptive mutation operator together to improve the genetic algorithm. The paper also introduces voting mechanism for it. Experimental results show that the algorithm performs better than Simple Genetic Algorithm(SGA), as it reduces the features and complexity of the Back Propogation Neural Network(BPNN) classifier apparently.

Key words: bi-directional selection, feature-fixing, feature selection, genetic algorithm, Back Propogation Neural Network(BPNN)

中图分类号: 

• TP311