论文部分内容阅读
已有的基于遗传算法的多标记特征选择算法成功地提高了多标记分类的准确率,但其求得高质量解的时间成本叫高,计算效率较低,基于此,提出了一种基于改进遗传算法与文化基因的多标记聚类方案。首先,基于特征与标记间的依赖将特征按照适应度进行排名,使用遗传搜索建立多标记类;然后,使用局部优化方案向被选的特征集中增加精英样本或删除较弱样本。对部分基因进行改良,且在搜索过程中逐渐改变优化操作的次数,从而降低了整体计算成本。最终,理论地分析了本算法的时间复杂度,获得了较好的时间效率。从多个角度进行了对比实验,结果表明本算法的求解质量与求解速度均优于其他多标记特征选择算法,本算法使种群始终保持较好的多样性,从而防止了早熟收敛。
The existing multi-label feature selection algorithm based on genetic algorithm has successfully improved the accuracy of multi-label classification, but the time cost of obtaining high-quality solution is called high, and the calculation efficiency is low. Based on this, Multi-label clustering scheme based on genetic algorithm and cultural gene. Firstly, the features are ranked according to their fitness based on the dependence of the features and markers, and the genetic markers are used to build the multi-marker class. Then, the local optimization scheme is used to add the elite samples to the selected feature sets or delete the weaker ones. Improve some of the genes, and gradually change the number of optimization operations in the search process, thus reducing the overall computational cost. Finally, the time complexity of this algorithm is theoretically analyzed and the better time efficiency is obtained. The results of experiments show that the quality of the proposed algorithm and its speed of solution are better than those of other multi-tag feature selection algorithms. This algorithm keeps the diversity of the population at all times and prevents premature convergence.