Cost-effective and scalable traffic forecasting: A graph-free multi-view MLP architecture

Area of Science:

• Intelligent Transportation Systems
• Urban Computing
• Machine Learning for Spatio-temporal Forecasting

Background:

• Current traffic prediction models face scalability issues due to Large Language Models' (LLMs) high costs and traditional methods' reliance on complex graph structures or attention mechanisms.
• Existing approaches struggle to balance prediction accuracy with the computational efficiency required for real-time, large-scale traffic operations.
• There is a critical need for lightweight, deployable predictive architectures in urban computing.

Purpose of the Study:

• To propose a novel, highly scalable, and graph-free computational framework for accurate traffic prediction.
• To address the deployment imperative by developing an efficient and lightweight predictive architecture.
• To overcome the limitations of existing spatio-temporal forecasting methods in real-world applications.

Main Methods:

• Introduced the Spatio-Temporal Multi-view MLP Network (STMMN), a graph-free framework.
• Utilized an adaptive spatial clustering multilayer perceptron (MLP) module to extract latent spatial homogeneity and streamline dimensionality.
• Developed a temporal multi-view aggregation module to fuse multi-scale temporal dependencies (proximity, trend, periodicity) without attention maps.

Main Results:

• STMMN demonstrated state-of-the-art prediction accuracy on multiple large-scale, real-world datasets.
• The proposed framework exhibited exceptional computational efficiency compared to existing methods.
• Achieved significant reduction in learning complexity by circumventing predefined topologies.

Conclusions:

• STMMN offers a highly scalable and computationally efficient solution for traffic prediction.
• The research validates the practical value of lightweight, knowledge-aware models for large-scale urban engineering.
• This work bridges the gap between theoretical accuracy and practical deployability in intelligent transportation systems.