<p>Network slicing has emerged as an effective solution for resource allocation in 5G networks, enabling the delivery of diverse services with distinct quality-of-service (QoS) requirements. This paper introduces a novel framework for predictive network slicing using an enhanced deep reinforcement learning algorithm, Deep Q-Network for Adaptive Slicing and Resource Allocation (DQN-ASRA). Leveraging a high-traffic event dataset from real 5G environments, the proposed model forecasts appropriate network slices based on traffic patterns and user behavior. The framework incorporates key enhancements like epsilon decay, reward shaping, prioritized experience replay, and regularization techniques to improve learning stability, convergence speed, and predictive accuracy. DQN-ASRA integrates slice prediction and dynamic resource allocation into a unified decision-making process, particularly targeting ultra-reliable low-latency communication (URLLC) scenarios. The model is trained and evaluated using performance metrics such as prediction accuracy, average reward, and training loss. Results indicate that the proposed approach delivers substantial improvements in both slicing accuracy and resource allocation efficiency under dynamic traffic conditions. Compared to baseline models, DQN-ASRA achieves a 7.7% increase in slicing accuracy, a 17.3% reduction in latency, and a 33% improvement in convergence speed. These findings demonstrate the effectiveness of advanced deep reinforcement learning techniques in solving complex problems in 5G network management. By adaptively responding to varying service demands, DQN-ASRA enhances end-to-end performance and contributes to dependable, low-latency 5G service delivery.</p>