Exploring the frontiers of trajectory outlier detection: an in-depth review and comparative analysis

Abstract

This paper provides a review and comparative analysis of trajectory outlier detection methods. It presents the definition of outliers in trajectory data and the existing types to further examine the advanced approaches. Basic steps for detecting an outlier, which include data preprocessing, feature extraction, modeling, and similar, have been presented. Moreover, advanced methods such as autoencoders and the use of deep learning for outlier detection have been explored. In the end, this paper evaluates the techniques and compares them using common metrics, mainly focusing on the techniques based on autoencoders or deep learning. It covers applications in real life and practice along with any limitations, challenges, and perspective ideas for the future. Ultimately, it can be a useful resource for expanding the understanding of domain researchers and practitioners.