Analyzing Key Factors Influencing Employee Resignation Through Decision Tree Modeling and Class Balancing Techniques

Abstract

Employee resignation poses a significant challenge to organizational stability and workforce planning. This study aims to analyze the key factors influencing employee resignation by developing an interpretable predictive model using the Decision Tree algorithm. The analysis is conducted on the IBM HR Analytics dataset, which includes 1,470 employee records with diverse demographic, behavioral, and job-related attributes. To address the issue of class imbalance—where resignation cases are underrepresented—the Synthetic Minority Over-sampling Technique (SMOTE) is applied to enhance model sensitivity and balance. After a comprehensive data preprocessing phase, including feature selection and label encoding, the Decision Tree model is trained with a limited depth to reduce overfitting and maintain interpretability. The model achieves an accuracy of 77%, with a recall of 0.80 and an F1-score of 0.77 for the resignation class. Feature importance analysis identifies stock option level, job satisfaction, monthly income, relationship satisfaction, and job involvement as the most influential predictors. These findings provide actionable insights for human resource practitioners seeking to implement targeted and data-driven employee retention strategies. The study highlights the practical value of interpretable machine learning models in human capital analytics.