Short-term Traffic Forecasting Using Graph Neural Networks on Taxi Data
Name
Pavlo Pyvovar
Abstract
Traffic forecasting (TF) is the task of predicting a traffic state in a city for a certain time horizon into the future. Accurate traffic forecasts are crucial for preventing congestion, which causes degradation of life quality for people, and improving mobility services that help citizens commute more efficiently. Machine learning methods have been applied extensively to TF problems and Graph Neural Networks (GNN) have recently become state-of-the-art methods. In this study, we aim to apply a GNN to a particular taxi probe dataset which was provided to us by an Estonian mobility company Bolt Technology OÜ. We show that our implementation of GNN is capable of learning seasonality from time series data and yet does not outperform traditional machine learning methods. So further improvements are needed to make a GNN that excels at TF.
Graduation Thesis language
English
Graduation Thesis type
Master - Computer Science
Supervisor(s)
Amnir Hadachi, Joonas Puura
Defence year
2024