Comparison of decision support models -- a recruitment case study as a multiclass classification problem with limited data

We present a comparative case study of machine learning models, evaluating their efficiency in a practical task of multiclass classification of samples being submissions to a recruitment survey and assigning them scores denoting the match level for a given candidate to a given workgroup (committee) in the AGH Students’ Council. This research is based on the Council’s recruitment applications that carried candidates’ responses to a set of 10 hypothetical Council member activity scenarios, where they were to choose one of four given solutions to the problems. The data was collected from a web quiz in 2020, validated on a voluntary insider control group’s responses to these questions and finally the best-performing model was evaluated in practice in the 2021’s recruitment inside an in-browser adventure minigame. This work provides insight into how models ranging from classical methods to deep learning perform in a very specific not yet well-explored in literature, practical non-linear problem that is dependent on individual features of the participants, with the data volume being very limited due to a restricted population of candidates. This information may provide a starting point for applications of machine learning in decision support systems in recruitment processes.