Studying Driving Factors of Employee Attrition Using Feature Importance Approaches

NIU, WENMIN (2020) Studying Driving Factors of Employee Attrition Using Feature Importance Approaches. [Dissertation (University of Nottingham only)]

[thumbnail of 20157952_BUSI4374 UNUK_Studying Driving Factors of Employee Attrition Using Feature Importance Approaches.pdf] PDF - Repository staff only - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader
Download (2MB)

Abstract

What factors affect employee attrition has been studied for many years. In machine learning filed, it is feature importance approach that helps researchers get insights about the driving factors in prediction problems. However, feature importance methods previous studies use are usually computed on individual models and they have obvious weaknesses. For example, permutation and “rebuilding” mask the importance of some features when correlation exists. One of the purposes of this dissertation is to uncover important features using novel MCR which give a comprehensive description of important features by studying how features are relied on by a set of “Rashomon” models. Another aim is to explore the exact relationship between important features and employee attrition possibility.

After comparing with “permutation”, MCR indicated that MCR+ gives more chance to features show their importance, so more features were deemed important, such as “EnvironmentSatisfaction”, “BusinesssTravel_Rarely”, “DistanceFromHome”, “MaritaStatus_Single”, “WorkLifeBalance” and “EducationField_Medical”. There are also insights for business and HR managers. For example, results from PDPs suggested that working overtime is a significant indicator of turnover. Satisfaction with job, environment and relationship is negatively related to attrition. Last but not least, stock options are very useful for improving employee attention, even if the lowest level of stock option can decrease the attrition possibility a lot, but increasing stock options does not help decrease the probability more.

Item Type: Dissertation (University of Nottingham only)
Depositing User: NIU, Wenmin
Date Deposited: 14 Dec 2022 09:20
Last Modified: 14 Dec 2022 09:20
URI: https://eprints.nottingham.ac.uk/id/eprint/61728

Actions (Archive Staff Only)

Edit View Edit View