Machine Learning Techniques in Employee Churn Prediction

Abstract

An employee has one of the most important roles in a company’s continued operation. This study employed different machine-learning techniques to identify the best classifier for employee churn prediction. A total of 14,249 instances were secured of which 14,200 instances were used as a training set and identifying the two best classifiers. The remaining 49 instances with removed class labels were used as a test set. Seven classifiers were employed in this study, these are the trees.J48 (decision tree), trees.RandomForest, trees.RandomTree, the lBk (k-NN), Naïve Bayes, Logistic, and the Multilayer Perceptron. Parameter tuning was implemented for the J48 and the k-nearest neighbor. The result revealed that the trees. Random Forest (97.76%) and trees.J48 with a confidence factor of 0.25 (96.73%) have the two highest classification accuracy values. In the prediction phase, these two classifiers were used to predict the test set, where 40 instances are predicted to stay and 9 instances are predicted to leave in the actual dataset. On the other hand, the trees. Random Forest predicted 38 employees to stay and 11 to leave, while the trees.J48 with a confidence factor of 0.25 predicted 35 employees to stay and 14 as leaving. Implications are discussed.