On a clustering-based mining approach for spatially and temporally integrated traffic sub-area division

Traffic sub-area division plays an important role in traffic control and is a critical task for traffic system management and traffic network analysis. Most existing algorithms for traffic sub-area division are based on traffic road networks and face a significant challenge in dealing with complex and time-varying traffic network conditions. This paper proposes a clustering-based method for Spatially and Temporally integrated Traffic Sub-area Division, referred to as ST-TSD, which takes into account a complete spectrum of spatiotemporal trajectory information. ST-TSD determines not only a set of traffic sub-areas but also a time interval when these sub-areas are formed without user intervention. In this method, we first establish a discrete linear representation of trajectory points to generate a series of trajectory segments and transform them into multidimensional data points in Euclidean space. We then design an algorithm to extract potential intensive time intervals based on multidimensional data points and improve an existing density clustering algorithm to divide the whole traffic network at each corresponding intensive time interval into a set of sub-areas. Finally, we employ the Convey Hull algorithm to identify the boundaries of filtered sub-areas. For performance evaluation, we design a traffic sub-area division indicator, referred to as TSDI, as a performance metric by combining the WCSS indicator and the classical Davies–Bouldin index. Experimental results on real-life trajectory datasets illustrate that the proposed ST-TSD method significantly improves the quality of traffic sub-area division over existing methods.