The current air traffic control system tries to allocate as many flights as possible in a scenario that is expected to be time-efficient, cost-efficient, and safe. To guaranty these safety conditions, it is performed a cyclic process known as Demand-Capacity Balancing. During this process, a specialized Air Traffic Controller analyses the situations where the demand is over the capacity to identify the required corrective actions. These corrective actions are mostly in the form of regulations, and they are necessary to avoid overload during the day of operation. The task of declaring a regulation is complicated, very time-consuming, and based on the Air Traffic Controller's experience. A massive amount of information must be considered simultaneously, together with a risk maturation process because of the uncertainty and granularity in the information. This paper proposes and evaluates two Deep Learning models able to mimic the current procedure's behavior, and therefore, helping the specialized Air Traffic Controller to automatically detect the imbalances that will require regulation. Both models, one based on Convolutional Neural Networks, and the second one based on Recurrent Neural Networks, have demonstrated the potential to predict regulations, with an accuracy of 81.45% and 80.73% respectively over the entire MUAC region in 30-minute intervals. This accuracy can be increased by up to 91% by developing specialized models for each airspace sector. Additionally, we performed an in-depth analysis of the most relevant features using SHapley Additive exPlanations.