基于动态时间跨度与聚类差异指数的用户行为异常检测算法

Abnormal user behavior detection algorithm based on dynamic time span and cluster difference index

在保证实时性与模型的适应性的条件下对居家人士的行为进行分析,提出了一种基于动态时间跨度与聚类差异性指数的用户行为异常实时检测算法。该算法利用动态时间跨度与聚类差异性指数对实时数据流进行概念漂移检测,在数据流发生概念漂移的情况下,利用局部离群因子(LOF)来检测用户发生行为异常的时间点。通过动态时间跨度对分类模型不断更新,有效提升了模型的适用性。通过实验验证了该算法能够在保证实时性的情况下正确检测出概念漂移,并给出用户行为发生异常的时间点。该研究成果为实现智能家居环境下用户行为异常检测提供了新思路,可为居家人士提供有效服务和安全保证。

In order to analyze the behavior of residents under the condition of ensuring real-time performance and adaptability of the model,this paper proposes a real-time detection algorithm for abnormal user behavior based on dynamic time span and clustering difference index.The algorithm uses dynamic time span and cluster difference index to detect concept drift in real-time data streams,and uses local outlier factor(LOF)to detect the time points when users have abnormal behaviors when concept drift occurs in data streams.The classification model is continuously updated through the dynamic time span,which effectively improves the applicability of the model.Experimental results show that the algorithm can correctly detect concept drift while ensuring real-time performance,and give the time point when user behavior is abnormal.The research results of this paper provide new ideas for the realization of abnormal user behavior detection in the smart home environment,and can provide effective services and security guarantees for home people.