Time Series Forecasting. In Data Science & Engineering Master of Advanced Study (DSE MAS) Capstone Projects

From the onset of the COVID-19 pandemic, a dramatic change in traffic patterns has been observed across the country due to travel and other restrictions imposed by government agencies and health experts. The causes for these abrupt changes can be at least partially attributed to the severity of the pandemic, the widespread increase in remote work and online learning, business closures, etc. Considering these changes, we hypothesize that the performance of static time series models used for traffic forecasting will degrade beginning in early 2020. Dynamic models that do not rely solely on historical information will better forecast day-to-day traffic and be able to learn long-term changes in traffic patterns. We present a graph convolutional recurrent neural network that captures both the inherent spatial and temporal complexities present in traffic forecasting. The algorithm is part of a larger deep learning library for time series modeling being developed for the open-source community.