Bicycle transportation, a low-carbon option, is essential for promoting sustainable urban mobility. However, predicting bicycle traffic is challenging due to limited investments in data collection, especially in smaller cities. This paper proposes a multi-source transfer learning spatial-temporal graph neural network (Multi-TLSTGCN) for accurate bicycle traffic prediction in target cities with limited available data. This study first examines how to transfer knowledge from single source domain to the target domain while mitigating the risk of negative transfer. Following this, a multi-source adaptive transfer learning approach is developed to optimize traffic prediction in the target domain by adaptively integrating knowledge from multiple sources. Finally, the performance of the Multi-TLSTGCN model is evaluated under various levels of target data scarcity and compared with models that do not incorporate source domain knowledge. The experimental results demonstrate several key insights: 1) Models fine-tuned with a single-cluster pre-trained source model where the clusters are formed based on similar traffic patterns are more effective at minimizing negative knowledge transfer than those fine-tuned with single-city pre-trained source models. 2) The proposed Multi-TLSTGCN outperforms baseline models in bicycle traffic prediction, showing promise for accurate predictions in data-scarce environments; and 3) The Multi-TLSTGCN model remains robust across varying levels of data scarcity, exhibiting only a slight decrease in accuracy as the availability of target data decreases, in contrast to models relying solely on target domain data. These findings highlight the Multi-TLSTGCN model as an effective and promising solution for bicycle traffic prediction with limited data availability.

Accurate short-term predictions of active mode traffic are crucial for effective urban traffic control and management, helping to reduce delays, stops, and improve travel time reliability, and optimize travel route choice. While most methods focus on motorized traffic, active modes like walking and cycling have been overlooked due to their complex dynamics and sensitivity to external factors like weather and individual choices, making them inherently less predictable. To address this, we propose a Dynamic Attention-based Spatial-Temporal Graph Convolutional Network (DyASTGCN) model that incorporates the impact of weather on graph spatial correlations within the active mode traffic network. Additionally, we introduce a fusion approach to integrate various heterogeneous spatial correlations, aiming to represent the optimal spatial correlations within the active mode network. Experimental results demonstrate that weather changes have a lagging effect on traffic network spatial correlations. Specifically, active mode traffic demonstrates significant sensitivity to precipitation, with notable changes in spatial correlations occurring within 5 minutes. Conversely, it takes approximately 20 minutes for spatial correlations to respond to wind speed influences. By incorporating both precipitation and wind speed with a 20-minute lag, our model outperforms those using only one feature, achieving the best traffic prediction performance. Given the uncertain traffic state and highly sparse nature of active mode data, our fusion approach adeptly captures the essential spatial correlations required for accurate traffic flow prediction. This allows our model to better understand complex graph correlations and traffic patterns, improving prediction accuracy and offering valuable insights into active mode network dynamics.

Accurate prediction of active mode traffic is imperative for optimizing traffic operations in Intelligent Trans-portation Systems. However, existing data-driven approaches heavily rely on extensive datasets to achieve reliable traffic prediction. This dependence poses a challenge when it comes to data sharing, particularly when collecting information from multiple local clients, such as institutions, organizations, and mobile devices, and transmitting it to a central server for model training and application. To overcome this challenge and enhance data security, we introduce the FedASTGNN model for active mode traffic prediction. This approach combines the federated averaging (FedAvg) algorithm with an attention-based spatial-temporal graph neural network (ASTGNN) model. Subsequently, we conduct an evaluation to determine the performance gap between the centralized ASTGNN model and the proposed distributed FedASTGNN model. This evaluation takes into account the model's performance across different time aggregation intervals and prediction horizons. Moreover, considering the unique attributes and intricacies of active mode data, we create three scenarios to demonstrate the influence of diverse active mode data from different local clients (subnet-works) on the FedASTGNN model. The findings of our study illustrate that the FedASTGNN model effectively preserves the advantages of the ASTGNN model while ensuring data confidentiality in active mode traffic prediction. Furthermore, we observe that the performance of the FedASTGNN model is significantly affected by the varying degrees of imbalanced data distribution among subnetworks. The insights shed light on the potential and challenges presented by the FedASTGNN model as an efficient and secure solution for predicting active mode traffic in Intelligent Transportation Systems.