Abstract

Nowadays the incredibly advanced developments in information technologies have led to exponential growth in the datasets with respect to both the dimensionality and the sample size. This trend can be easily illustrated in popular online data repositories (e.g., UCI machine learning repository). The more growth in the datasets, the more challenging the data management becomes. This is because such datasets usually comprise a high level of noise as well as the necessary information. Therefore, the elimination of noisy features in the datasets is an important task to discover meaningful knowledge. Although a large number of feature selection approaches have been proposed in the literature to deal with noisy features, the need for the studies based on feature selection has not come to an end. In this paper, we propose differential evolution-based feature selection approaches wrapped around the principles of fuzzy rough set theory. In contrast to well-known feature selection criteria such as standard mutual information, standard rough set and probabilistic rough set, our approaches can also deal with real-valued variables without the requirement of discretization. Moreover, the feature subsets selected by our approaches can profoundly improve the classification performance compared to the recent particle swarm optimization approaches based on probabilistic rough set and the state-of-the-art filter approaches.

中文翻译：

---

基于微分进化和模糊粗糙集理论的特征选择新滤波方法

摘要

如今，信息技术的惊人发展已导致数据集在维数和样本量方面均呈指数增长。在流行的在线数据存储库（例如，UCI机器学习存储库）中可以轻松说明这种趋势。数据集的增长越多，数据管理就越具有挑战性。这是因为此类数据集通常包含高水平的噪声以及必要的信息。因此，消除数据集中的噪声特征是发现有意义的知识的重要任务。尽管在文献中已经提出了大量特征选择方法来处理嘈杂的特征，但是基于特征选择的研究的需求还没有结束。在本文中，我们提出了围绕模糊粗糙集理论原理的基于差分进化的特征选择方法。与众所周知的特征选择标准（例如标准互信息，标准粗糙集和概率粗糙集）相比，我们的方法还可以处理实值变量，而无需离散化。此外，与最近的基于概率粗糙集和最新过滤器方法的粒子群优化方法相比，我们的方法选择的特征子集可以极大地改善分类性能。我们的方法还可以处理实值变量，而无需离散化。此外，与最近的基于概率粗糙集和最新过滤器方法的粒子群优化方法相比，我们的方法选择的特征子集可以极大地改善分类性能。我们的方法还可以处理实值变量，而无需离散化。此外，与最近的基于概率粗糙集和最新过滤器方法的粒子群优化方法相比，我们的方法选择的特征子集可以极大地改善分类性能。