Modeling the Probabilities of Vehicle Crashes Using Machine Learning Methods with Congestion Amounts and Spatiotemporal Features at Montgomery County, Maryland

Keywords: traffic congestion, time-series analysis, GeoAI, machine learning, transportation
Abstract Type: Paper Abstract

Abstract

Predicting traffic congestion and vehicle crashes at a regional scale helps choose driving routes, develop transportation plans, and design intelligent transportation systems. It also helps in the effective use of resources such as policing and facility installations. It, however, has not been studied comprehensively because of the limitation of congestion data resources available and appropriate metrics to summarize congestion. However, recent advancements in technology, such as the distanceTime congestion unit and Google Traffic Layer, have opened up new opportunities for research using big data. Furthermore, the development of machine learning methods provides a new approach to modeling congestion and crashes. The study focuses on understanding the relationship between traffic and crashes in a heavily populated suburban area. Specifically, the objectives are 1) forecasting traffic congestion amounts, 2) modeling crashes using congestion and spatiotemporal features, and 3) analyzing geographical patterns of potential crashes. Montgomery County, Maryland was selected as a study area for the analysis, and six-month traffic data were collected. The Long Short-Term Memory (LSTM) model was used to forecast time-series traffic volumes. After forecasting, its results and the spatiotemporal features are used as input to predict vehicle crashes in the study area. In this analytical step, we considered three different ways: 1) traditional statistics-based method, 2) machine learning, and 3) deep learning. The results of this study showed that machine learning methods can effectively predict traffic amounts and probabilities of crashes with spatiotemporal measurements. The findings of this research have the potential to be integrated into intelligent traffic systems.