In the upcoming years, en route airspace capacity will be limited by air traffic controller workload, requiring the introduction of automation to assist controllers with conflict detection and resolution. However, acceptance is considered to be one of the main obstacles in the introduction of novel automation. Individual-sensitive automation has been proposed to increase acceptance by adapting to different controller strategies. This research evaluates how personalized automation for air traffic control can be achieved using convolutional neural networks. A human-in-the-loop experiment is devised to generate datasets consisting of conflict resolution commands with corresponding velocity obstacle images as learning feature. Results show that the trained models can reasonably predict command type, direction and directional value. Furthermore, a correlation is found between a controller consistency metric and achieved prediction performance. Finally, the individual-sensitive models performed significantly better than general group-based models, confirming the strategy heterogeneity of the population, which is a critical assumption for personalized automation.