This study aims to analyze low bid deviations in transportation infrastructure projects, identify key project and market-related factors influencing these deviations, and employ a machine learning (ML) framework to improve bid accuracy and forecasting.
This study analyzes 2,915 historical transportation project bids from Louisiana (2012–2024). The dataset was split into training (80%) and testing (20%) subsets. A two-stage feature selection process was implemented, combining Pearson correlation filtering and mutual information ranking to select the top nine predictive variables. Five ML models, LightGBM, CatBoost, XGBoost, TabNet, and a stacking ensemble, were trained and evaluated. Model interpretability was enhanced through SHapley Additive exPlanations (SHAP) and sensitivity analysis.
TabNet achieved the best predictive performance, capturing complex, nonlinear relationships among variables. SHAP analysis revealed that the number of bidders, construction investment levels, and material costs were the most influential factors. Sensitivity analysis confirmed the measurable impact of construction demand and input prices on predicted deviations, while employment-related variables showed lower marginal influence.
This study is limited by its reliance on historical bid data from a single U.S. state (Louisiana), which may constrain the generalizability of the findings to other regions with different procurement frameworks, contractor behaviors, and economic dynamics. Furthermore, while the selected ML models demonstrated strong predictive performance under typical market conditions, their effectiveness may decrease during periods of economic volatility or unexpected disruption. These limitations highlight the need for future research to incorporate multi-state datasets, consider temporal modeling strategies, and explore regional heterogeneity in contractor markets and procurement practices. Such expansions would help validate and enhance the robustness of ML-based frameworks in broader public infrastructure contexts.
The results underscore the practical value of using ML-based tools, such as TabNet, in transportation cost estimation and procurement planning. By identifying key predictors of bid deviations—such as competition levels, construction demand, and material prices—transportation agencies can refine early-stage cost estimates and reduce estimation bias. Additionally, using interpretable models with SHAP and sensitivity analysis empowers decision-makers with actionable insights, promoting transparency and informed procurement strategies. Encouraging broader contractor participation and integrating real-time market indicators into cost estimation workflows can ultimately enhance project planning accuracy and procurement efficiency.
This research presents a replicable, interpretable ML framework for analyzing low bid deviations in public infrastructure projects. The approach enables transportation agencies to better understand the drivers of bid variability, improve cost estimation practices, and adapt procurement strategies to evolving market conditions.
