Predictive modeling of workplace accident outcomes utilizing XGBoost and Tree-Structured Parzen Estimator
Loading...
URL
Journal Title
Journal ISSN
Volume Title
School of Business |
Master's thesis
Unless otherwise stated, all rights belong to the author. You may download, display and print this publication for Your own personal use. Commercial use is prohibited.
Authors
Date
2022
Department
Major/Subject
Mcode
Degree programme
Business analytics
Language
en
Pages
48+17
Series
Abstract
Workplace accidents induce a cost of hundreds of millions of euros for insurance companies annually and indirectly even higher costs for employers and society, including human suffering. However, most of these costs are driven by the employee’s recovery time, and by advancing employees returning to work, the cost of workplace accidents can be reduced. From the perspective of insurance companies, employees’ return to work can be advanced by helping the accident victim get appropriate care as soon as possible by improving the administrative process. One way to advance the administrative process is by prioritizing cases with a higher risk of prolonged absence from work. Therefore, this research aimed to develop a prediction model to identify victims of workplace accidents that are likely to suffer a prolonged absence from work to help direct resources where most needed. The seriousness of the accident was defined based on the absence from work, and accidents, where the absence was more than 30 days, were considered serious. Following this, a binary classification model was developed utilizing ESAW variables described in the accident notice submitted to the insurance company. The model used in this research was XGBoost, and it was optimized using Tree-Structured Parzen Estimator (TPE). In addition, the model was trained using accident notifications delivered to insurance companies and collected by the Finnish Workers Compensation Center. The model could predict serious accidents with an accuracy of 73% and non-serious accidents with an accuracy of 77%.Description
Thesis advisor
Malo, PekkaKeywords
XGboost, TPE, accident outcome prediction, ESAW