Trajectory analysis using data mining techniques

Abstract

This study presents an innovative method for identifying variability in vehicular flow, designed for dynamic
urban environments with fluctuating traffic conditions. The proposed approach integrates real-time data flow
processing with a two-level clustering strategy to detect and analyze vehicular density patterns. The first level
performs dynamic clustering of GPS locations, forming microclusters that represent spatially homogeneous
traffic zones. Each microcluster is continuously updated based on similarity criteria and a forgetting mechanism
that ensures data relevance. Periodic snapshots capture the temporal evolution of the traffic distribution, which
serves as input for the second level of clustering. The second level aggregates microclusters based on proximity,
taking advantage of historical density data to classify traffic variability. By comparing current and baseline
densities, the method identifies congestion-prone areas and dynamically adjusts cluster formations. This twolevel
approach improves traffic management and provides a robust framework for detecting congestion trends.
Through validation in three urban case studies, San Francisco, Rome and Guayaquil, the methodology
successfully captured the spatial and temporal variability of traffic, identifying congestion hotspots and
uncovering patterns of flow evolution over time.