HLGF-Stack: A Risk-Aware Hybrid Ensemble Framework with Structured Human-in-the-Loop Governance for Enterprise Workflow Automation

Abstract The increasing adoption of intelligent automation in enterprise workflow systems has significantly transformed operational decision-making, resource allocation, and performance optimization. However, the growing dependence on machine learning–based delay prediction raises concerns regarding reliability, governance transparency, and operational risk exposure. Predicting workflow delays in dynamic enterprise environments remains challenging due to heterogeneous task attributes, evolving constraints, and the need for accountable automation.This paper proposes a hybrid ensemble framework with structured human-in-the-loop governance to enhance predictive robustness and enterprise accountability. The HLGF-Stack framework integrates Random Forest, Gradient Boosting, XGBoost, Support Vector Machine, and K-nearest neighbors within a stacked meta-learning architecture. Beyond predictive modeling, the framework incorporates feature engineering, calibrated probability estimation, and a composite risk scoring mechanism that evaluates predictive uncertainty, business criticality, and compliance sensitivity. A governance controller dynamically determines whether decisions are automatically executed or escalated for human review, ensuring a balanced trade-off between automation efficiency and oversight control. Experimental evaluation on the Workflow Operations Performance Dataset, characterized by significant class imbalance, demonstrates 98.73% accuracy, 97.53% precision, 98.69% recall, and a 98.75% F1-score. Governance analysis further reports 92.76% escalation precision and a Governance Efficiency Index of 0.461. The results confirm that integrating ensemble intelligence with structured governance significantly enhances reliability, transparency, and deployment readiness in enterprise workflow automation.