基于优化野草算法的加权模糊粗糙特征选择研究

发布时间：2018-05-01 01:01

本文选题：特征选择 + 模糊集　；参考：《大连海事大学》2017年硕士论文

【摘要】：特征选择是一种用来降低数据集维度的技术,其核心是从输入的特征集合中选择出最具有预测性的特征子集来代表原始数据集合。特征选择不仅可以简化特征内在的关系还可以改善整体集合的预测能力。目前,许多学者针对模糊粗糙集的特征选择进行了大量的研究,其中比较常见的有,遗传算法、蚁群算法(ACO)、粒子群算法(PSO)等。这些算法在鲁棒性和求解能力等方面均表现优秀,并且它们的共同特点是只有最优秀的个体才能有机会被提取出来。然而,某些初始依赖度值低的个体有可能带有重要信息,因此上述这些算法可能会导致重要信息丢失。针对以上问题,本文研究了野草算法的特点并且发现其特点能使模糊粗糙特征选择更加全面。野草算法认为初始依赖度值低的个体有可能带有重要信息,因而赋予初始依赖度值低的个体一定的存活机会。该算法早期能够维持特征种群多样性而在后期能够保证最优解的选择优势。因此,本文首先将野草算法的特点和模糊粗糙集理论相结合,继而提出了基于优化野草算法的加权模糊粗糙特征选择算法并对其进行编程实现。其次,利用基于模糊粗糙集的快速属性约简算法来验证特征选择结果。最后,将算法模型应用于十四类基准数据集和四类具有现实背景意义的乳腺造影数据集进行特征选择,并且将本文算法的特征选择结果与其他两个算法(蚁群算法和粒子群算法)的特征选择结果分别从分类精度和AUC值两个方面做出对比分析。数据分析结果表明,基于本文算法得到的大部分特征选择结果可以很好地代表原始数据集并且整体性能优于蚁群算法和粒子群算法。同时,这也证明了本文算法具有现实研究意义。
[Abstract]:Feature selection is a technique to reduce the dimension of data set. The core of feature selection is to select the most predictive feature subset from the input feature set to represent the original data set. Feature selection not only simplifies the intrinsic relationship of features, but also improves the prediction ability of the whole set. At present, many scholars have done a lot of research on the feature selection of fuzzy rough sets, among which the common ones are genetic algorithm, ant colony algorithm (ACOO), particle swarm optimization (PSO) and so on. These algorithms are excellent in robustness and solving ability, and their common feature is that only the best individual can be extracted. However, some individuals with low initial dependency may have important information, so these algorithms may lead to the loss of important information. In view of the above problems, this paper studies the characteristics of the weed algorithm and finds that its characteristics can make the fuzzy rough feature selection more comprehensive. The weed algorithm considers that the individuals with low initial dependency may have important information, so the individuals with low initial dependency are given a certain chance of survival. The algorithm can maintain the diversity of characteristic populations in the early stage and ensure the selection advantage of the optimal solution in the later stage. Therefore, this paper first combines the characteristics of weed algorithm with fuzzy rough set theory, and then proposes a weighted fuzzy rough feature selection algorithm based on optimal weed algorithm and implements it by programming. Secondly, the fast attribute reduction algorithm based on fuzzy rough set is used to verify the feature selection results. Finally, the algorithm model is applied to 14 kinds of datum data sets and 4 kinds of mammography data sets with realistic background significance for feature selection. The result of feature selection of this algorithm is compared with that of other two algorithms (ant colony algorithm and particle swarm optimization algorithm) from classification accuracy and AUC value. The results of data analysis show that most of the feature selection results based on this algorithm can represent the original data set well and the overall performance is better than that of ant colony algorithm and particle swarm optimization algorithm. At the same time, it also proves that this algorithm has practical significance.
【学位授予单位】：大连海事大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP18

【参考文献】