Graph based traffic analysis and delay prediction

Abstract

This study addresses the issue of traffic congestion in Malta, which is a common problem faced by many developed countries. In order to tackle this issue, the study proposes the collection and publication of a comprehensive traffic dataset, allowing individuals to experiment and analyse such data, given that any local traffic data is either private or shared in an aggregated, statistical report dating back to the previous year. This custom-built dataset includes realistic trips made by members of the public across the island known as MalTra. A methodology is also used to generate syntactic data to generate a complete dataset, allowing for a complete dataset to be created, and ensuring that the data is reliable and comprehensive. However, while traffic prediction in Malta is possible, an additional volume and a wider diversity of data would be beneficial for a more accurate result. It is worth mentioning that Malta has a rapid daily growth in vehicles, with around 11,000 cars being added between the first and last day of data collection. This indicates the importance of having an accurate and comprehensive means of collecting data to tackle the issue of fluctuating traffic in Malta. To analyse this dataset, several approaches are taken, including the use of graph theory, temporal-spatial graphs and deep learning models, namely ARIMA, DCRNN, and STGCN models. This research also compares the performance of two deep learning models, STGCN and DCRNN, where overall, the DCRNN model, consistently outperforms STGCN., obtaining an MEA score of 3.98 and 6.65, respectively. In conclusion, we propose a traffic dataset, MalTra, that provides individuals with access to reliable and comprehensive data, allowing for a better understanding of traffic patterns in Malta. This research compares the performance of two deep learning models, STGCN and DCRNN, where overall, the DCRNN model, consistently outperforms STGCN, obtaining an MEA score of 3.98 and 6.65, respectively. However, more data would be beneficial for more accurate traffic predictions, considering the rapid daily growth in vehicles in Malta.