Adaptive Data Rate is a feature of LoRaWAN that allows to optimize the network performance by adjusting the data rate of end devices based on their current channel conditions. Current approaches to ADR optimization algorithms focus on static or low-mobile end devices, and the specification recommends to disable ADR for mobile devices, e.g. location trackers. To let these devices also benefit from ADR adjustments, this thesis suggests to implement a predictive ADR algorithm based on deep reinforcement learning. The algorithm is evaluated on a real-world data set captured in the city of Darmstadt.