IoT networks are often composed of spatially distributed nodes. This is why mobile data collection (MDC) emerged as an efficient solution to gather data from IoT networks that tolerate delay. In this paper, we study the use of reinforcement learning (RL) to plan the data collection trajectory of a mobile node (MN) in cluster-based IoT networks. Most of the existing solutions use static methods. However, in a context where the MN has little information (no previous data set) about the environment and where the environment is subject to changes (cluster mobility, etc.), we want the MN to learn an energyefficient trajectory and adapt the trajectory to the significant changes in the environment. For that purpose, we will train two reinforcement learning (RL) algorithms: Q-learning and state-action-reward-state-action (SARSA) combined with deep learning (DL). This solution will allow us to maximize the collected data while minimizing the energy consumption of the MN. These algorithms will also adapt the trajectory of the MN to the significant changes in the environment.