This project was funded by the National Institute for Transportation and Communities (NITC; grant number 1299), a U.S. DOT University Transportation Center.
Machine learning, Transportation -- Safey measures, Bicycle transportation
This project focuses on giving bicyclists a safer and more efficient path through a city’s signalized intersections. It builds on a prior NITC project that tested an app for a fixed-time corridor. The goal of this project is to lay the groundwork for extending this earlier app to include actuated signals. Two machine-learning algorithms are introduced that have a good track record with time-series forecasting: LSTM and 1D CNN. The algorithms are tested on data captured from a busy bike corridor on the south end of the University of Oregon campus. A specific actuated intersection is identified on this corridor and real-time data is collected from it. The algorithms are trained on the data and evaluated. The results show that both algorithms can reach 85% accuracy and can predict on a single sample within roughly one second. While these results are encouraging in terms of adding a prediction component to the existing app, a closer look at Precision and Recall is more mixed. A means of computing a Precision-Recall tradeoff is discussed.
Fickas, Stephen. Green Waves, Machine Learning, and Predictive Analytics: Making Streets Better for People on Bikes, NITC-1299. Portland, OR: Transportation Research and Education Center (TREC), 2021. https://dx.doi.org/10.15760/trec.264