| 面向特征选择问题的协同演化方法 |
| |
| 引用本文： | 滕旭阳,董红斌,孙静.面向特征选择问题的协同演化方法[J].智能系统学报,2017,12(1):24-31. |
| |
| 作者姓名： | 滕旭阳 董红斌 孙静 |
| |
| 作者单位： | 哈尔滨工程大学 计算机科学与技术学院, 黑龙江 哈尔滨 150001 |
| |
| 摘 要： | 特征选择技术是机器学习和数据挖掘任务的关键预处理技术。传统贪婪式特征选择方法仅考虑本轮最佳特征，从而导致获取的特征子集仅为局部最优，无法获得最优或者近似最优的特征集合。进化搜索方式则有效地对特征空间进行搜索，然而不同的进化算法在搜索过程中存在自身的局限。本文吸取遗传算法（GA）和粒子群优化算法（PSO）的进化优势，以信息熵度量为评价，通过协同演化的方式获取最终特征子集。并提出适用于特征选择问题特有的比特率交叉算子和信息交换策略。实验结果显示，遗传算法和粒子群协同进化（GA-PSO）在进化搜索特征子集的能力和具体分类学习任务上都优于单独的演化搜索方式。进化搜索提供的组合判断能力优于贪婪式特征选择方法。
|
| 关 键 词： | 特征选择 遗传算法 粒子群优化 协同演化 比特率交叉 |
Co-evolutionary algorithm for feature selection |
| |
| TENG Xuyang,DONG Hongbin,SUN Jing.Co-evolutionary algorithm for feature selection[J].CAAL Transactions on Intelligent Systems,2017,12(1):24-31. |
| |
| Authors: | TENG Xuyang DONG Hongbin SUN Jing |
| |
| Affiliation: | College of Computer Science and Technology, Harbin Engineering University, Harbin 150001, China |
| |
| Abstract: | Feature selection is a key preprocessing technology of machine learning and data mining. The traditional greed type of feature selection methods only considers the best feature of the current round, thereby leading to the feature subset that is only locally optimal. Realizing an optimal or nearly optimal feature set is difficult. Evolutionary search means can effectively search for a feature space, but different evolutionary algorithms have their own limitations in search processes. The evolutionary advantages of genetic algorithms (GA) and particle swarm optimization (PSO) are absorbed in this study. The final feature subset is obtained by co-evolution, with the information entropy measure as an assessment function. A specific bit rate cross operator and an information exchange strategy applicable for a feature selection problem are proposed. The experimental results show that the co-evolutionary method (GA-PSO) is superior to the single evolutionary search method in the search ability of the feature subsets and classification learning. In conclusion, the ability of combined evaluation, which is provided by an evolutionary search, is better than that of the traditional greedy feature selection method. |
| |
| Keywords: | feature selection genetic algorithm (GA) particle swarm optimization (PSO) co-evolution bit rate cross |
|
| 点击此处可从《智能系统学报》浏览原始摘要信息 |
|
点击此处可从《智能系统学报》下载全文 |
