This thesis set out to investigate whether it is possible to apply machine learning to subjective employee evaluations. This was done through a scoping review to identify dominant issues in subjective evaluation of people and through a series of experiments using employee data to train machine learning algorithms. The scoping review reviled that the dominant issue in subjective evaluations are leniency bias which were accompanied by severity and centrality bias. This was used to focus the experiments on a specific problem, reducing or removing leniency bias from the evaluation. In order to test whether this was possible, 5 algorithms were trained on synthetic employee data. Neither of the algorithms performed better than random, which indicates that training an algorithm based on that dataset was not possible. During the training, it was apparent that the algorithms learned the human biases and were therefor not a possible solution. It was also evident that the dataset was skewed, which can have caused the bad performance of the algorithms. This means that it is not possible to make algorithmic evaluations of employees by applying a large dataset. It also requires HR-departments to have enough knowledge of measuring and performance to generate the right data for the algorithms to be trained by.
|Educations||MSc in Human Resource Management, (Graduate Programme) Final Thesis|
|Number of pages||106|