摘要： 军事训练领域的特殊性造成其相关数据存在大量的噪声点，同时也为噪声检测算法提出了相应的要求。分析现有数据噪声点检测算法，提出将数据属性分为空间属性、环境属性、特征属性，利用空间属性确定数据对象的分类，利用环境属性确定具有不同特征的数据对象邻域，利用特征属性计算离群度。改进了基于相对密度的离群度计算方法，提出LRDF算法，实验结果表明，该方法有效地提高了噪声点检测的精度和效率，增强算法可用性。

关键词: 军事训练数据, 数据挖掘, 噪声点检测, 相对密度, 属性分类

Abstract: The particularity of military training causes the outlier phenomenon of the data in this domain, and also brings special demand of the outlier detection method. Some existing representative algorithms have their localization in this domain. In general, the attributes of data object are classified as the spatial attributes, the environment attributes and the characteristic attributes. The spatial attributes categorize data objects, the environment attributes decide the neighborhood data objects field, and the characteristic attributes count the outlier score. On the bases of the improvement of relative density method, this paper proposes a new detection algorithm based on Local Relative Density Factor(LRDF), and experimental result shows that this algorithm improves the detection accuracy, efficiency and usability compared with the other existing algorithms.

Key words: military training data, data mining, outlier detection, relative density, attributes classification

中图分类号: 

• TP311