Integrated density and buffer-based stream data clustering for detecting anomalies in the internet of moving things

Abstract

Detecting anomalies in roadway environments is a vital application of intelligent transportation systems (ITS). It involves setting up sufficient sensing systems, reading and analyzing generated stream data, and clustering the data. Typically, there are numerous sensors that are movable, which suggests high dimensionality of data with high dynamics. Clustering high-dimensional stream data with relatively high dynamics is not an easy task. Existing stream data clustering creates stages of buffering before enabling the production of legitimate clusters. However, none of the existing clustering methods are adapted to operate effectively for moving vehicles. In this study, we incorporate an offline phase for creating clusters based on the density of core mini-clusters in an existing buffer-based online clustering for evolving data streams (BOCEDAS). We designate it as multi-density data stream (MUDEDS) clustering. The goal is to detect and cluster anomalies in the roadway environment. In addition, we build simulations for various types of anomalies in the roadway environment. Experimental results demonstrate the superiority of MUDEDS when evaluated on passing traffic signal, accident, and slippage types of anomalies compared to benchmarking clustering algorithms.