Abstract: In order to solve the problems of high time complexity, low accuracy and low recall in detecting anomaly database attributes, this paper proposes a new method based on ABC (artificial bee colony) and the O-measure metric (i.e., a kind of attribute outlier evaluation metric) to find out the attribute outliers, which simulates the bee colony  behavior of searching for high quality food sources. In view of the local convergence of swarm intelligence algorithm to detect the attribute outliers, this paper presents the approach of finding the global optimal solution by using the simulated annealing technique, making the bee swarm jump out of the local optimal solution. The proposed algorithm calculates the O-measure of each attribute that the bees have walked, and then from the O-measure value result sets, chooses the best food sources (i.e., the attribute outliers). In comparison with other algorithms, the experimental results show that the proposed algorithm needs less time, and improves the detection precision and recall.

Key words: attribute outlier, artificial bee colony algorithm, simulated annealing, O-measure

摘要： 为了解决数据库属性异常点检测方法时间复杂度大并且查准率和查全率不高的问题，提出了新的基于人工蜂群优化技术（artificial bee colony，ABC）和O-measure度量（一种评估属性异常点的度量）相结合的属性异常点检测方法，模拟人工蜂群随机搜索较优的食物源能力发现属性异常点。针对群体智能算法检测属性异常点会陷入局部收敛的缺陷，提出使用模拟退火技术让人工蜂群跳出局部最优解而找到全局最优解的算法。该算法通过蜂群在二维数据平面上搜索食物源, 计算所经过路径上的数据项O-measure适应度，从中寻找最优解（即属性异常点）。实验结果表明，所提算法较之前的算法耗时短，且提高了检测的准确率和查全率。

关键词: 属性异常点, 人工蜂群算法, 模拟退火, O-measure