<p>With the rapid pace of urbanization, challenges such as traffic congestion and air quality deterioration have become increasingly pronounced, making traffic information forecasting a key component in addressing these challenges within intelligent transportation systems. However, traffic flow exhibits complex dependencies in the spatiotemporal dimensions, and the prediction difficulty is further exacerbated by external events such as traffic accidents. While current models based on neural networks have seen some success, their ability to express nonlinear patterns and provide interpretability remains limited. To this end, this paper presents a traffic information prediction model, Graph Convolutional Multivariate Kolmogorov–Arnold Network, which integrates the multivariate Kolmogorov–Arnold Network (KAN) with the Graph Convolutional Network (GCN). The model first extracts the spatial features of the traffic network through the GCN layer, capturing the topological dependency relationships between roads. Subsequently, it employs three KAN variants—Chebyshev KAN, B-spline KAN, and Taylor KAN—to handle periodicity, smoothness, and nonlinearity in traffic flow, respectively. By dynamically evaluating the characteristics of the traffic data and appropriately assigning weights to each variant, the model adaptively optimizes the capture of different features. The model’s clear mathematical formulation improves the clarity and interpretability of the forecasting process. Experimental results demonstrate that the proposed model outperforms traditional approaches in capturing complex spatiotemporal nonlinear relationships, significantly improving the accuracy and interpretability of traffic flow predictions. This research provides a novel solution for traffic information forecasting in Intelligent Transportation Systems.</p>