Predicting the Occurrence and Causes of Employee Turnover with Machine Learning

Abstract

This paper looks at the problem of employee turnover, which has considerable influence on organizational productivity and healthy working environments. Using a publicly available dataset, key factors capable of predicting employee churn are identified. Six machine learning algorithms including decision trees, random forests, naïve Bayes and multi-layer perceptron are used to predict employees who are prone to churn. A good level of predictive accuracy is observed, and a comparison is made with previous findings. It is found that while the simplest correlation and regression tree (CART) algorithm gives the best accuracy or F1-score, the alternating decision tree (ADT) gives the best area under the ROC curve. Rules extracted in the if-then form enable successful identification of the probable causes of churning.