Predicting Competitive Bidding Levels in Transportation Projects Using Explainable Machine Learning and Deep Learning Methods

Abstract

Abstract Maintaining robust competition in transportation construction bidding is challenging for State Departments of Transportation (DOTs) as they navigate fluctuating input prices, complex project scopes, and variable market conditions. Existing approaches to forecasting bidder participation often overlook the combined effects of project-level attributes, market dynamics, and pre-bid signals. The objective of this paper was to forecast bidder participation for Georgia transportation lettings using bid information and market context, enabling DOTs to make proactive, evidence-based bidding and scheduling decisions. This study assembles 2017–2024 letting records from Bid Express (BidEx) and augments them with state agency and federal economic, labor, and energy indicators. Competition is framed as a three-class target, Low (1–2 bidders), Medium (3–5), and High ( ≥ 6 ), and modeled using tree-based learners and deep neural networks to capture nonlinear interactions and heterogeneous effects in structured pre-bid and market data. All preprocessing, tuning, and threshold optimization are performed on the training split, with the held-out test set reserved for a single independent evaluation. Gradient boosting delivers the highest class-balanced performance [ Accuracy = 0.71 , Macro -averaged F1 score (Macro-F1) = 0.69 , area under the receiver operating characteristic curve ( AUC ) = 0.86 ], followed closely by random forest. Relative to deep learning models, the multilayer perceptron achieved 0.775 Accuracy and 0.743 Macro-F1, with strong recall in high-competition cases, and appears to outperform the gradient boosting machine (GBM) baseline. SHapley Additive exPlanations (SHAP) and permutation analyses confirmed that the number of bidders was the strongest predictor, followed by project size, district, and energy prices. The findings translate into actionable levers: expand bidders’ engagement to increase participation, right-size packages to deter entry, and schedule lettings to avoid months with elevated energy prices or overloaded portfolios. The results of this study enable DOTs to forecast bidder competition with High precision using only pre-bid data, thereby supporting timely, cost-effective project delivery.

The authors' abstract, as published at the source. Journal of Construction Engineering and Management, 2026 · DOI ↗