Leveraging Machine Learning for Forecasting Workforce Stability

Workforce stability is vital to an organization's sustainable performance, but many organizations still experience costly employee turnover, resulting in lost experience, time, and money. This study uses machine learning to predict workforce stability by changing the conventional problem of “who will leave” to a predictive problem of “which employee is most likely to stay or leave the organization, and why?” The IBM HR Employee Attrition dataset is used, which was cleaned, encoded, normalized, and balanced using the Synthetic Minority Over‑sampling Technique (SMOTE). After this, descriptive analysis and correlation analysis are undertaken to identify the most significant factors behind stability, primarily age, distance from home, monthly income, marital status, and gender. A voting meta-learner that combines Random Forest and Extreme Gradient Boosting (XGBoost) as base learners is then developed as an ensemble model. The ensemble achieves competitive performance accuracy 0.89, AUC 0.85, significantly outperforming Random Forest and performing comparably to XGBoost accuracy 0.88, AUC 0.82. However, minority-class recall remained moderate at 0.40, improving to 0.58 after threshold tuning, indicating that while the model is effective for overall classification, detecting at-risk employees remains challenging. The evaluation, encompassing accuracy, precision, recall, F1-score, and AUC, supports the practical utility of the proposed configuration as a decision-support tool for workforce stability forecasting. The empirical results indicate that economic and demographic factors, mainly monthly income and commuting distance, are most important for stability, while gender has a comparatively limited impact.