The performance of cellular networks depends heavily on accurate traffic prediction, as it is essential for smooth service delivery and effective resource management. However, the unpredictable and constantly changing patterns of network activity make this task highly challenging. Traditional forecasting approaches often fall short in representing these complex dependencies, leading to reduced prediction accuracy and inefficiency. To address this issue, the objective of this work is to design a prediction model that is both accurate and computationally efficient. In this paper, we have proposed a lightweight hybrid prediction model integrated with an Advanced Temporal model with Attention Mechanism (ATAM). The temporal attention mechanism enhances the model’s ability to focus on relevant sequential patterns, while the hybrid architecture ensures computational efficiency and scalability for real-time deployment. The model’s performance is evaluated through Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and paired t-test demonstrating significant improvements over traditional forecasting approaches. These findings highlight that combining temporal attention with lightweight architecture enhances predictive performance while maintaining efficiency. In conclusion, the ATAM-based framework offers a reliable solution for traffic prediction in 5G and beyond cellular networks. Further our proposed model can be extended to support adaptive resource allocation strategies, thereby enabling operators to optimize network quality of service while reducing computational overhead.