Drones have been gaining popularity due to their increased use in various areas, such as delivery of goods, surveillance, infrastructure inspection, and others. Meanwhile, these very advancements have led to growing concerns about their potential effects on different facets of people’s safety and security. This includes the risk of a drone malfunction, causing it to stray away and collide with another flying or static object, e.g. a plane or a random person; or even the possibility of hackers abusing it by deliberately introducing anomalies thereinto. This prompted a surge in research to address these concerns using modern Machine Learning (ML) methods. Our study focuses on anomaly detection in drones, based on various logs that they produce throughout their flights, in real time and a posteriori. We introduce Log-processing Ensemble-Based Approach for Drone Anomaly Detection (LEBADAD), a multilayered model architecture that operates on log message statuses, starting with a data preprocessing and encoding phases, followed by a deep neural network ensemble that learns the log message sequences and subsequently detects deviations based on specific criteria. We show the evaluation results on online datasets along with one that we collect from our own in-lab Parrot ANAFI AI drone. We discuss enhancements to LEBADAD using fuzzy logic to account for potentially overlapping log message statuses that reflect certain real-world scenarios. Finally, we expand on conducting further model fine-tuning as part of future direction.