Mobile Edge Computing (MEC) improves the performance of real-time services by positioning computational resources at the network edge, closer to users, thereby minimizing latency. However, the decentralized nature of edge computing makes its infrastructure more susceptible to cyber threats, underscoring the need for intrusion detection systems (IDS) to enhance security. This study analyzes task offload decision-making in a MEC environment, where tasks can be executed either on the edge cloud or on a mobile device. To this end, a hybrid model combining machine learning techniques (Support Vector Machine, Random Forest, Gradient Boosting) and a Deep Q-Network (DQN) neural network is proposed. Experimental results based on data from Shanghai and Guangzhou demonstrate notable improvements, with accuracy rates ranging between 67 and 99%.