Machine Learning and Biodata: Predicting Early Employee Flight During Preselection

Voluntary employee turnover has consistently focused on retention after selection as opposed to occurring during the preselection process. This is significant in the manufacturing industry where this business problem causes high cost and the risk of reducing company value. This exploratory descriptive analytical (EDA) study adds to the existing bodies of knowledge in both the human resources and business intelligence fields by exploring the ability of selected biodata variables to have the potential to predict which might be significant drivers of a candidate’s decision to terminate within the first three years of employment. Using primary data accessed from a United States’ southeastern manufacturing firm between 2008 and 2015, this study provided statistical evidence that relocation, site experience, and rehire status could be have the potential to identify employees who may voluntarily leave their organization within three years of hiring. The study showed how to use multiple EDA algorithms (Random Forest, gradient boosting, linear discriminant analysis, and classification trees) to explore and reduce the data to a more prepare it for use in future research. Random Forest analysis was determined to be the most accurate in determining cluster specificity and reduced false positive findings. Findings do indicate that for the studied organization, hiring locally people with previous site-based experience might reduce the potential for early voluntary terminations. Further research potential from the data source includes the use of regression and specifically, logistic regression, to determine more predictive results