Accurate traffic prediction is essential for effective traffic management and planning. However, traffic prediction models are challenged by various factors such as complex spatiotemporal dependencies in traffic data. Recently, researchers have explored the new approach known as stream analysis since it can continuously update models by capturing new behaviors from the traffic data stream. However, applying this approach specifically raises the question about the balance between model complexity and model flexibility for dynamic updates. ADRIP - Adaptive multi-agent system for DRIving behaviors Prediction proposed in [1], [2] has combined the dynamic clustering and the multi-agent system approach to solve this challenge. This system has been applied to predict traffic dynamics at the road segment level. In this paper, we aim to extend ADRIP to complete its functionality for traffic prediction at the network level. Experiments for multi-scale traffic data are conducted to compare extended ADRIP with well-known clustering models.