This research introduces the use of deep reinforcement learning (DRL) algorithms to enhance Mobile Edge Computing (MEC) task offloading through 5G networks. The improved DRL task offloading has been designed to minimize energy consumption, to balance the total number of tasks being processed on all edge servers, and to reduce average time required to complete a task via choosing whether to offload tasks to edge servers or execute tasks directly on User Equipments (UE). The task offloading is performed as a Markov Decision Process (MDP) and is accomplished using a Deep Q-Network (DQN). The DQN agent uses information about tasks and the network to make decisions regarding the offloading of tasks. Training stability and learning effectiveness are enhanced through the use of experience replay and a target network in the DQN architecture. When compared with execution of tasks without offloading, task latency is reduced by nearly 60%. In addition, the proposed approach increases CPU utilization on edge servers and yields greater energy savings. Evaluation and implementation of the proposed method has been completed on the Mininet emulator with traffic workloads. Although there are constraints, including dependence, on task profiles and a centralized control agent the experimental findings highlight the approachs scalability and robustness. Overall, the framework offers a practical way to address resource management challenges in evolving edge computing settings, leading to more effective and responsive mobile applications.