摘要
提出一种建立在局部最优基础上的动态集成选择算法,并从理论上对算法进行了分析.该算法首先在多个局部特征空间上构造最优集成,然后使用动态集成选择技术对未知样本进行识别.局部空间上的集成构造问题被转换为一个单目标优化问题,并使用多种群遗传算法进行了求解.基于UCI数据集的实验表明,相对于现有的动态分类器选择算法和动态集成选择算法,新算法能够取得更高的识别率.同时,相对于现有的动态集成选择算法,新算法构造的集成规模更小,识别速度更快.
A novel dynamic ensemble selection algorithm based on local optimal ensembles（LOEDES） is brought up,and it is analyzed in theory.LOEDES firstly constructs optimal ensembles in local regions of the feature space,and then dynamically selects them during the recognition phase.The construction problem is converted to a single-objective optimization problem,which is solved by a multi-population genetic algorithm.Experiments on some UCI datasets show that LOEDES can get higher recognition rates than current dynamic classifier selection algorithms and dynamic ensemble selection algorithms.What′s more,compared with current dynamic ensemble selection algorithms,LOEDES can construct smaller ensembles,which can get higher recognition speed.
出处
《小型微型计算机系统》
CSCD
北大核心
2011年第5期1005-1011,共7页
Journal of Chinese Computer Systems
