AN EMPLOYEE ATTRITION PREDICTION METHOD BASED ON SMOTE AND ADABOOST

Decreasing the employee attrition rate for enterprises requires the prediction of employee attrition. The employee attrition data is unbalanced because it contains many more separated employees (instances) than active employees. This paper proposes an employee attrition prediction method (called SMOTE-AdaBoost) based on SMOTE and AdaBoost for unbalanced data. First, an improved SMOTE algorithm is established to balance the data. The proposed SMOTE algorithm uses a new distance measure and a new synthetic instance-generating strategy for the employee attrition data that have both continuous and discrete features. Then, an ensemble model with decision trees as base learning models is established with the AdaBoost strategy for employee attrition prediction. The proposed method is verified on two employee attrition datasets. The experimental results indicate that the proposed SMOTE algorithm significantly improves the prediction performance of the model for separated employees. Moreover, the proposed SMOTE-AdaBoost algorithm obtains significantly better performance for employee attrition than several typical classification algorithms.