基于核主成分分析的异常轨迹检测方法

针对现有算法不能有效应用于多因素轨迹异常检测的问题，提出基于核主成分分析(KPCA)的异常轨迹检测方法。首先，为了改善轨迹特征提取的效果，采用KPCA对轨迹数据进行空间转换，将非线性空间转换到高维线性空间；其次，为了提高异常检测的准确率，采用一类支持向量机对轨迹特征数据进行无监督学习和预测；最终检测出具有异常行为的轨迹。采用大西洋飓风数据对算法进行测试，实验结果表明，该算法能够有效提取出轨迹特征，并且与同类算法相比，该算法在多因素轨迹异常检测方面具有更好的检测效果。

Trajectory outlier detection method based on kernel principal component analysis

In view of the fact that the existing algorithms cannot effectively be applied to multi-factor trajectory outlier detection, this paper proposed a new method named TOD-KPCA (Trajectory Outlier Detection method based on Kernel Principal Component Analysis). Firstly, in order to enhance the effect of trajectory feature extraction, the method used KPCA to do the space transformation for trajectories and converted nonlinear space to a high dimension linear space. Furthermore, in order to improve the accuracy of outlier detection, the method used one-class Support Vector Machine (SVM) to do unsupervised learning and prediction with trajectory feature data. Finally, the method detected those trajectories with abnormal behavior. The proposed algorithm was tested on the Atlantic hurricane data. The experimental results show that the proposed algorithm can effectively extract trajectory features, and compared with the same algorithm, the proposed algorithm has better detection results in terms of multi-factor trajectory outlier detection.