Distributed unmanned systems are increasingly finding use in a variety of applications, viz. reconnaissance, disaster management, search and rescue, etc. Sensing and actuation are key for the correct operation of such systems, especially since they do not have centralized control. These types of distributed systems have shown to be susceptible to spoofing attacks, such as false data injection attacks (on sensor values) -either via Man-inthe-middle (MitM) mechanisms or counterfeit signal generation. While machine learning techniques have been used to detect anomalous behavior, it has not found use in this domain. In this paper we pose the following questions: (a) "how well does a feed-forward deep learning model perform in detecting sensor anomalies?" and (b) "how can we inflate the dataset to reduce the cost of collecting data?" The second question aims to assist with the process of generating data for training the deep learning models and we study the effectiveness of Generative Adversarial Networks (GANs) for this purpose. Using software-in-the-loop (SITL) simulations, we analyze the feasibility of learning the behavior an unmanned autonomous vehicle (UAV). We present our findings for both of these questions in this paper.