Non-recurrent congestion analysis using data-driven spatiotemporal approach for information construction

• We perform a data-driven method for Incident-Induced Delay on freeway corridor.
• We apply spatiotemporal clustering analysis for secondary incident and IID estimation.
• We develop a pattern matching algorithm to estimate the recurrent congestion.
• The algorithm is transferable to any similar system for incident management.

A systematic approach to quantify Incident-Induced Delay (IID) is proposed in this study. The paper complements existing literature by developing a data-driven method to dynamically determine the spatiotemporal extent of individual incidents. The information construction process can be further used to uncover a variety of features that are associated with any specific incidents for optimal freeway management. Additionally, this study contributes two particular highlights: secondary incident identification and K-Nearest Neighbor (KNN) pattern matching. Secondary incident identification, as a pre-processing for IID estimation, disentangles the convoluted influences of subsequent incidents. The proposed method uses KNN pattern matching, an essentially heuristic search process to separate the delay solely induced by incidents from the recurrent congestion. The proposed algorithm on IID quantification was implemented on Interstate 15 in the state of Utah using data obtained from 2013. Results and implications are presented. Hot spot analysis is conducted that can be potentially used for incident mitigation and to inform investment decisions. The proposed methodology is easily transferable to any traffic operation system that has access to sensor data at a corridor level.