摘要
针对现有异常轨迹检测中分类不平衡造成难以确定最优分类面的问题,提出一种基于加权极限学习机(ELM,Extreme Learning Machine)的异常轨迹检测算法。该算法采用加权ELM克服轨迹数据不平衡造成的分类面偏移,通过对正、负两类样本合理分配权重,并构造最优分类面获得较好的异常检测效果。仿真实验表明,加权ELM算法在训练速度,准确率,整体性能等方面均优于传统SVM和BP网络分类方法。
It is difficuh to find the optimal separating hyperplane caused by imbMance classification of the existing trajectory outlier detection algorithm, this paper proposes an algorithm to detect trajectory outliers by means of weighted extreme learning machine （ELM）. This algorithm adopts the Weighted ELM to overcome the offset of separating hyperplane. Firstly, proper weight is set for positive and negative samples adaptively, and then the optimal separating hyperplane is constructed to get better effect for abnormal detection. The results of simulation experiments show that, in training speed, accuracy and overall performance, the weighted ELM algorithm is better than the traditional SVM and BP network classification method.
出处
《微处理机》
2014年第1期76-79,84,共5页
Microprocessors
关键词
异常检测
迹轨分析
极限学习机
Outliers detection
Trajectory analysis
Extreme Learning Machine
