The study explores the use of HR data to predict employee turnover using machine learning models.
The goal of the analysis was to identify key factors influencing employee retention and turnover, comparing multiple machine learning models to determine the best fit.
Top important variables:
Satisfaction level, Time spend company, Work accident, Salary, Number project
Top important variables:
Satisfaction level, Number project, Average monthly hours, Time spend company, Last evaluation
Top important variables: Satisfaction level, Time spend company, Salary, Work accident, average montly hours
Top important variables: Satisfaction level, Number project, Last evaluation, Time spend company, Average monthly hours
Top important variables: Time spend company, Number project, Last evaluation, Satisfaction level, Average montly hours
Considering the company’s perspective, we created a new metric called "Total Gain" by combining profit and loss hypotheses from the company's perspective.
profit=0.4*PPV+0.4*TPR+0.2*Accuracy
Loss=0.5×error+0.4×FNR+0.1×FPR
Total Gain=Profit-Loss
When the model was fitted using Random Forest, the prediction success rate was as high as 98.8%, and the error rate for remaining (0) in terms of whether or not to change jobs (Left) was close to 0%.
Taking into account the company's perspective, we added a new metric called "Total Gain." Our research showed that employee satisfaction, tenure, and average monthly working hours had a bigger impact on employee attrition than salary.
Our results revealed that the Random Forest model consistently outperformed other models, highlighting "employee satisfaction" as the most influential variable affecting turnover.
The distribution of "average monthly working hours" for employees who left the company was spread across both low and high hours, and it appeared correlated with the number of projects. Employees with fewer working hours likely handled fewer projects, while those with longer hours managed more. It can be inferred that those with high workloads left due to pressure, while those with lower workloads left due to low achievement. Surprisingly, salary had less impact on turnover than expected, indicating that factors other than salary play a larger role in deciding whether to stay or leave.