基于分组规划的特征选择研究

With the continuous development of science and technology, the ability of data collection and storage has rapidly developed and the era of data explosion has arrived. The dimensionality reduction of high dimensional data has become the research focus of many scholars. Feature selection is one of the most effective and popular methods for dimensionality reduction. Especially, for high dimensional data, dimensionality reduction shows more significant advantages. Recognizing the essence of feature and seeking a fast and stable feature selection method with high precision for high-dimensional data is still an unresolved problem. Completely different with the existing methods, the idea of our work starts from the existing metastable state of "intrinsic" structure of optimal feature subset, studying the composition of the structure, the formation of laws, constraints, building a theoretical model (group planning model) interpretation, characterization, identifying optimal feature subset, in order to provide direction guide and theoretical support for the feature selection. Based on the results of the aforementioned theories and methods, we plan to make use of Bacterial Foraging Optimization(BFO) and Fuzzy Logic Control to build a new and fast feature selection method with high precision and suitable for high dimensional data, and further use of the feature clustering and multi-task learning to adapt to the lack of classification label data, unbalanced data, thereby inhence the stability of feature selection method. This project has great significance in promoting the theory of feature selection and dimensionality reduction method for high dimensional data.

随着科学技术的不断发展，数据的采集和存贮能力迅猛发展，数据爆炸的时代已经来临，高维数据降维成为众多学者的研究热点。特征选择是降维最有效最普遍的方法之一，尤其针对高维数据降维呈现出更为显著的优势。认清特征选择本质并为实际应用寻求快速、高精度、稳定的特征选择方法仍是一个未被解决的难题。与现有方法的研究思路完全不同，我们从最优特征子集“内在”结构存在的亚稳定性出发，研究结构的组成、形成规律、约束条件，构建一种理论模型（分组规划模型），解释、刻画、识别最优特征子集，从而为特征选择提供方向引导和理论支持。在以上理论和方法的研究结果上，利用细菌觅食优化算法（BFO）和模糊逻辑控制构建快速、高精度的高维数据特征选择新方法，进一步利用特征聚类和多任务学习适应缺失分类标签数据、不平衡数据，从而增强特征选择算法的稳定性。本项目对促进特征选择理论及高维数据降维方法的研究具有十分重要的意义。

随着科学技术的不断发展，数据的采集和存贮能力迅猛发展，数据爆炸的时代已经来临，高维数据降维成为众多学者的研究热点。特征选择是降维最有效最普遍的方法之一，尤其针对高维数据降维呈现出更为显著的优势。认清特征选择本质并为实际应用寻求快速、高精度、稳定的特征选择方法仍是一个未被解决的难题。与现有方法的研究思路完全不同，我们从最优特征子集“内在”结构存在的亚稳定性出发，研究结构的组成、形成规律、约束条件，构建一种理论模型，解释、刻画、识别最优特征子集，从而为特征选择提供方向引导和理论支持。主要研究内容包括：特征选择的理论模型研究、特征选择新方法、稳定特征选择方法、优化算法、机器学习共性关键技术、分类器优化方法。上述研究提高了特征子集的质量，可有效提高分类器的稳定性、精度、数据适应性。本项目的成果已应用于多个在研项目，可一进步应用于数据挖掘、计算机视觉、无人系统等领域。