In case of unexpected or temporary events, cellular networks can become quickly saturated. A promising solution is using drones as flying base stations. In this article, we address the issue of anomalous behavior within cellular networks that occurs during crowded events. The proposed approach consists of two parts: the detection of overloaded cells using a machine learning algorithm (LSTM - Long Short-Term Memory) and the deployment of drone base stations to assist the cellular network by providing wireless coverage. Initially, we use the LSTM algorithm to analyze the impact of extra-data on the network and detect the peaks of user's demands. Then, we formulate an optimization problem for maximizing the coverage when deploying drones taking into account the energy constraints. The proposed approach is validated using real data-set extracted from the Call Detail Records (CDR) of Milan. Simulation results show that the use of drones can satisfy the quality-of-service requirements of the network.