Employee Turnover Prediction on Synthetic and Real Datasets

When an employee leaves the job, it leads to issues such as financial losses, loss of productivity, re-employment costs, the adaptation period for the new employee, and loss of time. Many studies are dedicated to predicting employee turnover to avoid these negative impacts. In this study, we train classic machine learning models to predict employee turnover. We train Logistic Regression, Decision Tree, XGBoost, Random Forest, Gradient Boosting, Ada Boosting, and Support Vector Machine methods. We use two publicly available datasets: IBM's synthetic data and an anonymous US company's real data. We compare the outcomes of models trained by collected real datasets and models trained by synthetically generated datasets. For the IBM dataset, with 92% accuracy and 0.92 f1-score, Random Forest demonstrates superior results. Additionally, Random Forest out- performed other models with 89% accuracy and 0.89 f1-score for the US company's dataset. All applied classification models demonstrated better results on the synthetic data.