Optimising Construction Contract Rates with Z-Score and Machine Learning Approaches

Abstract

This study evaluates two common techniques for contracting rate rationalisation in construction using Z-score statistical screening and machine learning (ML) models on a real-world tender dataset (single project; ten tenderers). We compare a human-in-the-loop, two-stage process where Z-scores act as an initial, auditable filter to detect unusual bids, and ML (regression and clustering) develops context-aware price benchmarks from the multivariate features. Empirically, the ML method achieved a 38% reduction in the RMSE compared to a Z- score-only baseline, whereas the hybrid approach reduced false alarms and produced more consistent and clearer pricing ranges for decision support. Z-score screening maintained operational value by standardising anomaly detection and enhancing procedural transparency, whereas ML improved the predictive accuracy and adaptability to item, project, and market differences. The results show that combining transparent statistical rules with data-driven models improves the accuracy, efficiency, and fairness of tender assessments. The contributions of this study are threefold: (i) a direct comparison of Z-score versus ML at the Bill-of-Quantities level; (ii) a governance-ready hybrid protocol that blends auditable thresholds with model-based benchmarks; and (iii) practical guidance for agencies seeking to embed analytics within e-procurement systems. Although the scope is limited to a single-project dataset, the findings support further validation across multiple projects and real- time deployment with drift monitoring in future research.