This study reconceptualizes quality management in project-based construction as a governance mechanism rather than a reactive inspection function. It shows how interpretable Machine Learning (ML) can operationalize TQM into transparent quality gates regulating cost escalation, project risk and audit attention under construction data constraints.
A staged mixed-methods design was used. Practitioner interviews and expert consensus defined governance-critical variables, which were embedded into a synthetic dataset for simulation-based internal evaluation. Multiple linear regression, logistic regression and Isolation Forest models represented rework-cost signaling, risk escalation, and anomaly-based audit functions. Robustness testing and a two-round Delphi study assessed structural coherence and practical relevance.
The framework showed strong internal coherence. Rework-cost estimation explained substantial cost exposure (R2 = 0.8156), identifying training coverage and defect accumulation as dominant governance levers. Risk classification achieved high discrimination (AUC = 0.9425), and anomaly detection identified unusual projects for targeted review and learning. The framework provides a structured, interpretable decision-support architecture for quality planning and escalation control.
Once calibrated with project-level quality data, the framework can convert fragmented quality records into governance signals for prioritizing quality investment, escalating high-risk projects and targeting audits before quality failures become costly.
Rather than proposing new algorithms, this study contributes a simulation-evaluated decision-rule architecture linking rework-cost signaling, risk escalation gates and anomaly-triggered audit learning within one governance workflow, advancing Quality 4.0 toward interpretable, governance-compatible decision support.
