Hierarchical Workforce Dynamics and Risk-Aware Gated Deep Learning Framework for Employee Attrition Prediction

Human resource management is a key factor in organizational success. Understanding and predicting employee attrition is important for improving workforce planning and decision-making. In this study, we propose a Hierarchical Workforce Dynamics – Temporal Risk-Aware Gated model named HWD-TRAG deep learning framework for accurate and transparent attrition prediction. The proposed model integrates hierarchical learning, risk-aware adaptive gating, and adaptive feature relevance scaling to capture complex workforce patterns. It learns important employee factors such as workload, job satisfaction, income, promotion delay, and work-life balance in a structured way. The model also considers how employee behavior changes over time, making it more realistic for real-world scenarios. To ensure transparency, SHAP and attention-based fusion are used to explain model predictions clearly. Experimental results show that the proposed model achieves a high accuracy of 97.1%, outperforming all baseline methods. For explainability, the model achieves a faithfulness score of 0.95, explanation consistency of 0.96, and feature stability of 0.94, while maintaining low sparsity of 0.22. In terms of robustness, it achieves strong performance with noise robustness of 0.95, data shift robustness of 0.94, and feature perturbation stability of 0.96, which confirms the proposed model's performance.

@article{Yang2026,
  title     = {Hierarchical Workforce Dynamics and Risk-Aware Gated Deep Learning Framework for Employee Attrition Prediction},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {17},
  number    = {7},
  year      = {2026},
  publisher = {The Science and Information Organization},
  author    = {Chen Yang},
  doi       = {10.14569/IJACSA.2026.0170761},
  url       = {https://doi.org/10.14569/IJACSA.2026.0170761}
}