A high drop-out rate is present during current-day air traffic controller (ATCo) training, because the required expertise level is not reached. The determination of the expertise level of ATCo students is currently performed using subjective assessments at a late stage in the training by means of high-fidelity simulator sessions. It is desired to objectively measure expertise earlier and more frequently in training to monitor the progress of the student. However, it is currently unknown which objective measures might describe the expertise level of an ATCo. This paper presents a method that identifies a set of objective measures that can classify an ATCo's expertise level using a genetic algorithm and hierarchical agglomerative clustering. A large set of possible objective measures and a dataset containing data from 10 ATCos (intermediate and pro level) is used. The method found a set of 8 measures that can cluster the 10 ATCo's in the two expertise groups very accurately (97,5% accuracy). The genetic algorithm showed a preference for measures that have a distinction in the results between the expertise groups. However, not all selected measures show a difference between the expertise groups, resulting in signs of overfitting. Furthermore, these measures only provided limited feedback for the individual ATCos. Clustering the results of the 10 ATCo's gave valuable information about the overall expertise level of an ATCo, as compared to the average intermediate- or pro-ATCo.