This paper presents a machine learning approach to the aircraft and crew recovery problem. The presented model utilises a two-stage sequential approach recovering the aircraft and flight schedule first, followed by that of cockpit crew. At each stage, the recovery is assisted via use of decision tree-based machine learning classifiers to select a subset of relevant aircraft and crew, reducing the scope of the recovery operation. The aircraft stage is able to directly account for aircraft maintenance constraints, and indirectly account for the impact of the aircraft recovery on crew schedules and passenger itineraries. In the crew stage, the model directly accounts for the common cockpit crew labour constraints set by airlines operating within the US. The combined performance of the recovery model is evaluated via a case study on the US network of Delta Airlines. The results show that the crew selection algorithm can find an optimal solution to the crew recovery problem in 89% of non-trivial disruption instances, completing in 27 s on average, with an average objective function value only 5% higher than optimal. When including aircraft recovery results, the proposed approach increased the percentage of instances that obtain a solution within the AOCC-defined time limit from 47% to 95%, speeding up the average computational time threefold. The solution to the aircraft and crew recovery problem was globally optimal in 83% of cases, with an average objective function value 11% higher than optimal.