Context‐Aware Network Traffic Prediction and Resource Allocation: An AI‐Enabled Network Slicing for Smart City

Abstract

Smart cities represent a critical vertical use case that demands comprehensive attention in managing heterogeneous, temporal, and dynamically managed traffic demands. This dynamic management of smart city faces two primary challenges: (1) identification of AI‐driven approaches capable of managing heterogeneous, complex, and scalable smart city data, and (2) enabling efficient management and reallocation of network resources during sudden fluctuations in traffic demands. To address these challenges, this work first proposed a hybrid model that combines LSTM and Q‐learning for traffic prediction and resource optimization. Subsequently, LLM is incorporated to handle contextual information and dynamic resource reallocation in response to sudden fluctuations in traffic demand. Experimental results show that the proposed model achieves a prediction accuracy of 95%, a Normalized Mean Absolute Error (NMAE) of 0.032, and a Pearson's Correlation Coefficient (PCC) of 0.92, indicating a strong alignment between predicted and actual traffic loads. In addition, the latency variability is reduced by 15%, and the RL agent attains a cumulative reward of 250.5. The integrated LSTM‐Q‐learning‐LLM framework achieves a classification accuracy of 97.3%, demonstrating superior performance in handling dynamic network traffic scenarios. Analysis reveals that the proposed model outperforms individual LSTM and Q‐learning models, showing a 10% improvement in prediction accuracy, a 5% greater reduction in latency variability, and a higher cumulative reward during training. This work provides a solid foundation for the development of adaptive and intelligent network management systems in smart city environments.