Predicting employee attrition with machine learning on an individual level, and the effects it could have on an organization

Abstract: This paper is investigating the possibility to predict employee attrition on an individual level with machine learning. The study is divided into two parts, one qualitative part which were conducted by doing interviews with selected roles where the openness to which practitioners are willing to use machine learning models to predict employee attrition, and what effects such a model could have on an organization was investigated. The second part is a quantitative part where a random forest model, support vector machine model and a logistic regression model are compared in terms of accuracy in predicting employee attrition with the usage of large human resource data sets. Firstly, it was shown that people are willing to use machine learning models to predict employee attrition if the models were to be trusted, and if organizations that used such models were transparent in how the models were used, and to what purpose. The model comparison did not give any interesting results about the possibility to predict employee attrition with the chosen models. There were several reasons for that, where some of them were that the models were over fitted, the time of notice when a person quit was not accounted for enough and the choice of input data points. This resulted in that the accuracy could not be determined in a confident way.