Predicting voluntary employee turnover using core employee data

Predicting voluntary employee turnover using core employee data

van Bekhoven, Sjoerd (TU Delft Electrical Engineering, Mathematics and Computer Science)

Loog, Marco (mentor)
Reinders, Marcel (graduation committee)
Kouw, Wouter (graduation committee)
Jonker, D. (graduation committee)
Bhulai, S (graduation committee)

Delft University of Technology

2017-08-30

Voluntary employee turnover is the process of an employee voluntarily choosing to resign from a company. High voluntary turnover has been shown to have negative effect on both organisational and financial performance of companies. Therefore, if companies were to know which individuals would leave their company in the coming months, this would open doors for many applications in human resources (HR) man- agement. However, data directly related to voluntary employee turnover (for example: exit interviews) is scarce, while all companies have HR information systems in which they capture core employee data. This study therefore aims to research to what extent voluntary employee turnover can be predicted on an individual employee level using only core employee data. We design and extract features related to voluntary turnover using the hierarchical structure of the company such as features measuring team diversity and the hierarchical position of an employee in the company. This allows us to considerably increase performance compared to only using the original numerical features. Despite this increase, we feel performance needs another boost to make it usable in practice. We therefore investigate the underlying problems, showing that the root problem is not likely to be the inherent class imbalance or a lack of sample size, but rather a severe class overlap, meaning that the extracted feature sets are likely not informative enough to separate the voluntary leavers from the other employees.

voluntary employee turnover
people analytics
binary classification
feature design
class imbalance
class overlap
feature importance

http://resolver.tudelft.nl/uuid:95411815-890a-4aba-b6e0-336c9080cfc0

Student theses

master thesis

© 2017 Sjoerd van Bekhoven