Abstract
Building a transportation network robust enough to support ferrying a city’s large population between the various areas of interest is a hard problem, and traffic congestion and management are common challenges faced by every metropolis. Despite its well-developed public transportation system, Singapore is no exception, facing high levels of congestion with adverse effects on its society and environment. Efforts to mitigate such challenges in private transportation in Singapore have been made by both the government and private companies, with technology being at the core of these innovations. However, traffic decisions are still largely made by individuals in a user-centric way. For achieving system-wide travel time optimality, centrally coordinated mobility on demand (MoD) and autonomous mobility on demand (AMoD) vehicles need access to current and predicted traffic travel times. With advancements in machine learning and deep learning technology, graphical convolutional networks (GCN) have arisen as the state-of-the-art traffic travel time prediction method due to their ability to model the graph-like structure of a road network’s contextual information that would affect the prediction for the target road. Specifically, we make use of the Spatio-Temporal Graph Convolutional Network (ST-GCN) to capture both temporal and spatial features of the road network in Singapore for traffic travel time prediction. The contributions of our work include: (a) a contextual ST-GCN model which uses external contextual information and a new weighted loss function for a 3% improvement in short-term predictions over the original ST-GCN model and a 1% improvement overall across all output timesteps, (b) historical traffic travel time data collection and processing to train the contextual ST-GCN prediction model and (c) an open sourcing real-time traffic travel time prediction API. In addition, we also include a political, economic, social, technological, legal, and environmental (PESTLE) analysis of the Singapore market and businesses, a market size estimate based on ride-hailing services in the country, and a comparison to existing systems using the blue ocean strategy framework. Findings conclude that Singapore is a suitable market to support smart transportation technologies with a sufficiently large market, good technological infrastructure, and strong support from the government. Finally, various valid growth strategies for the system are discussed in the form of an Ansoff matrix with the ideal of creating fully-fledged MoD and AMoD system in Singapore.