Employee turnover prediction using machine learning based methods

Kısaoğlu, Zehra Özge
Employee turnover is a major problem for many companies because it brings with new issues including hiring costs, overtime costs, low productivity. Hence, preventing or reducing turnovers is a challenging task in human resource management field. At this point, employee turnover prediction plays an important role in providing early information for highly probable turnovers in near future that enables companies to take precautions against this situation. In this thesis, we work on this problem to predict whether an employee will leave his/her company within a certain time period. We formulate this binary classification problem as a supervised machine learning problem. Our study exploits publicly available employee profiles taken from the Web and job transition graphs extracted from these profiles. Main contribution of our study on predicting turnovers is the forming and use of job transitions of employees as well as the publicly available information about employees and institutions. So far, most of the turnover prediction models are built with the statistical methods or data analysis techniques and they make use of detailed employee information like age, race, job performance in company or job satisfaction survey results. To the best of our knowledge, this is the first study on predicting turnovers using job transitions and machine learning methods. With the help of job transition graph analysis and relevant features extracted from the graphs, many machine learning models under the change of year and time period parameters are composed. Several experiments with several models on different years’ employee profiles indicate that our proposed models have considerably predictive capabilities compared to different baselines.
Citation Formats
Z. Ö. Kısaoğlu, “Employee turnover prediction using machine learning based methods,” M.S. - Master of Science, Middle East Technical University, 2014.