A-Gamba: An Adaptive Graph-Mamba Model for Traffic Prediction in Wireless Cellular Networks

Abstract

Accurate traffic prediction is essential for optimizing network resource allocation in wireless cellular networks. In this letter, we propose an adaptive graph-Mamba (A-Gamba) model for traffic prediction. The proposed model utilizes a bidirectional-Mamba block and a selective scan algorithm to capture the temporal dependencies by prioritizing relevant historical data. We develop an adaptive graph convolutional block that captures the spatial dependencies between cells by learning a graph structure without requiring the location information. Results show that our proposed A-Gamba model outperforms three state-of-the-art baselines, achieving a lower root mean squared error and mean absolute error with faster training time.