IEEE Trans Cybern. 2022 Jun;52(6):4508-4519. doi: 10.1109/TCYB.2020.3029338. Epub 2022 Jun 16.
This article introduces a new model to identify a group of trajectory outliers from a large trajectory database and proposes several algorithms. These can be split into three categories: 1) algorithms based on data mining and knowledge discovery, which study the different correlations among the trajectory data and identify the group of abnormal trajectories from the knowledge extracted; 2) algorithms based on machine learning and computational intelligence methods, which use the ensemble learning and metaheuristics to find the group of trajectory outliers; and 3) an algorithm exploring the convolution deep neural network that learns the different features of historical data to determine the group of trajectory outliers. Experiments on different trajectory databases have been carried out to investigate the proposed algorithms. The results show that the deep learning solution outperforms data mining, machine learning, and computational intelligence solutions, as well as state-of-the-art solutions in terms of runtime and accuracy performance.
本文提出了一种新的模型，用于从大型轨迹数据库中识别一组轨迹异常值，并提出了几种算法。这些算法可以分为三类：1）基于数据挖掘和知识发现的算法，研究轨迹数据之间的不同相关性，并从提取的知识中识别出异常轨迹组；2）基于机器学习和计算智能方法的算法，使用集成学习和元启发式算法来找到轨迹异常值组；3）探索卷积神经网络的算法，该算法学习历史数据的不同特征，以确定轨迹异常值组。在不同的轨迹数据库上进行了实验，以研究所提出的算法。结果表明，在运行时间和准确性方面，深度学习解决方案优于数据挖掘、机器学习和计算智能解决方案，以及最新的解决方案。
