基于竞争策略的链式智能体遗传算法用于特征选择的研究

Link-like Agent Genetic Algorithm for Feature Selection Based on Competition Strategy

针对特征选择问题,提出了基于竞争策略的链式智能体遗传算法(LAGA)。该LAGA算法包含链式智能体网络结构,邻域竞争,自适应交叉,自适应变异,优良个体替换策略,自适应结束等部分,该算法能较好的保持智能体的多样性,在进化中既较佳的继承了优良个体的基因,又有效地搜索了新的空间。多组实验结果表明,通过该算法选择得到的最优特征子集具有较好的稳定性,较高的识别准确率和较低的网络分类器维数复杂度。

According to feature selection problem, a new algorithm to do feature selection is proposed based on linklike agent genetic algorithm （LAGA）. This algorithm includes link-like agent structure, neighborhood competition, adaptive crossover, adaptive mutation, replacement strategy, and adaptive stopping criteria. It can keep the diversity of the agents well, and effectively inherit the good genes of good individuals as well as search new space. Empirical results show that the feature subset obtained through the algorithm has better stability, higher classification rate and lower dimensional complexity of NN classifier.