针对现有基于密度的孤立点检测算法的不足,给出了一种新的基于密度的孤立点定义,并提出了相应的基于双半径(DR)密度差异的孤立点检测(OD)算法——DROD。该算法通过考察数据空间内任一点的双邻域半径与邻域半径内的数据密度的差异,有效地识别孤立点。DROD算法摒弃了传统孤立点检测方法中的k-近邻查询,大大节省了时间开销。在人工数据集和真实数据集上的实验表明,与现有孤立点检测算法相比,DROD算法在时间复杂度和孤立点的寻找能力方面均有较优表现。
In view of the drawbacks of existing density-based outlier detection algorithms, the paper gives a new density-based outlier definition, and proposes a corresponding algorithm for outlier detection (OD) based on density difference of double radius (DR) DROD. The DROD compares difference density of double radius and radius for each point in feature space in order to detect oufliers effectively. Without the time-consuming k-NN queries as in traditional outlier detection methods, the algorithm reduces the time complexity greatly. The results of the experiment on both synthetic datasets and real datasets show that the algorithm has better performance on both time complexity and the ability to detect outliers.