Early Prediction of Employee Turnover Using Machine Learning Algorithms

Abstract

Employee turnover is a serious challenge for organizations and companies. Thus, the prediction of employee turnover is a vital issue in all organizations and companies. The present work proposes prediction models for predicting the turnover intentions of workers during the recruitment process. The proposed models are based on k-nearest neighbors (KNN) and random forests (RF) machine learning algorithms. The models use the dataset of employee turnover created by IBM. The used dataset includes the most essential features, which are considered during the recruitment process of the employee and may lead to turnover. These features are salary, age, distance from home, marital status, and gender. The KNN-based model exhibited better performance in terms of accuracy, precision, F-score, specificity (SP), and false-positive rate (FPR) in comparison to the RF-based model. The models predict the average probability percentage of turnover intentions of the workers. Therefore, the models can be used to aid the human resource managers to make precautionary decisions; whether the candidate employee is likely to stay or leave the job, depending on the given relevant information about the candidate employee.