Accurate cost forecasting in budget planning and contract bidding is crucial for the success of construction projects. Linear models such as the autoregressive integrated moving average (ARIMA) and nonlinear models such as the artificial neural network (ANN) have been adopted in the literature for forecasting construction costs. However, both linear and nonlinear models are subject to some limitations derived from their modeling structure and assumptions. This study proposes a hybrid ARIMA-ANN model for forecasting construction costs and explores whether the hybrid ARIMA-ANN model can provide more accurate forecasts than an individual ARIMA or ANN. The national and city-level construction cost indices (CCIs) are forecasted for three forecasting horizons (short-term, mid-term, and long-term) using three forecasting models: (1) linear ARIMA, (2) nonlinear ANNs, and (3) the hybrid ARIMA-ANN model. Out-of-sample forecasting exercise reveals that the hybrid model combining the distinctive features of both ARIMA and ANNs performs better than individual models in most forecasting cases, especially for longer-term forecasting horizons. The findings can help project planners, cost engineers, and decision makers prepare for more accurate budgets and bids for diverse construction projects in different locations.

Because of the importance of budget planning and contract bidding, accurate prediction of future movements of the construction cost index (CCI) has been a crucial part of construction cost management. Despite the fact that construction costs can vary widely across locations with different market conditions and environments, the national CCI, the simple average of construction costs for 20 US metropolitan areas, has often been used to forecast CCIs across the nation. This study finds considerable differences across US cities in both the level and growth rates of CCIs and shows that using the national CCI for predicting city-level CCIs can cause nonnegligible forecast errors. Comparing four popular linear forecasting models, including standard autoregressive integrated moving average (ARIMA) models and a multivariate vector error correction model (VECM) drawn on monthly CCI data from January 1995 to December 2019, this study reveals that no single model is dominant in terms of the out-of-sample performance. If any leading indicator is available at the city level, however, it is recommended to choose between the ARIMA model based on the Bayesian information criterion (BIC) rule and the multivariate VECM augmented with the leading indicators. For the leading indicator in multivariate VECM, city consumer price index (CPI) works better than national CPI. In the absence of such leading indicators, it is recommended to use a parsimonious ARIMA model based on the BIC rule. The proposed approach is expected to help decision makers in the construction industry prepare more accurate budgets and biddings for regional construction projects.

Given the importance of accurate contingency estimation to properly absorb cost fluctuations, there have been enormous efforts to identify risk factors and measure contingency in roadway projects. However, no study has proposed a contingency framework that reflects the linkage effects of factors through the project phases. To this end, this research develops a project phase-based contingency adjustment framework using a path analysis drawing on a dataset of 1,053 rehabilitation, reconstruction, and resurfacing (3R) roadway projects in Florida. In particular, our framework incorporates the sequential effects of factors through the lenses of project phases from planning and bidding and contract to construction and completion. The developed path model shows that individual factors may have direct and/or indirect impacts on project costs. The results of this study can be applied to better contingency and cost estimation and proper risk management in the preconstruction phases of roadway renewal projects.

Large language models (LLMs) have achieved impressive results across various tasks. However, they still struggle with long-context documents. This study evaluates the performance of three leading LLMs: GPT-4o, Claude 3.5 Sonnet, and Gemini 1.5 Pro on lengthy, complex, and opinion-varying documents concerning infrastructure projects, under both zero-shot and few-shot scenarios. Our results indicate that GPT-4o excels in zero-shot scenarios for simpler, shorter documents, while Claude 3.5 Sonnet surpasses GPT-4o in handling more complex, sentiment-fluctuating opinions. In few-shot scenarios, Claude 3.5 Sonnet outperforms overall, while GPT-4o shows greater stability as the number of demonstrations increases.

Free-form concrete panel (FCP) molds require precise manufacturing because each mold demands a unique shape. Therefore, automation technology for producing these molds is being developed. However, when concrete is cast in a free-form mold and subjected to the impact of compaction to distribute it, deformation occurs in the precisely designed form. Consequently, free-form molds are often produced manually, which results in bugholes on the surface of FCPs. These bugholes lead to quality issues in the panels, including deterioration in aesthetics and strength. This study aims to develop a magnetic compaction technology that installs an object which rotates due to magnetic force inside a free-form mold and applies magnetic force from the bottom to perform compaction according to the free form. By comparing a control group using the existing manual FCP production method and an experimental group using magnetic compaction, strength was measured and bughole incidence was verified. As a result, although the experimental group was subjected to rotary motion, no material separation or deterioration in strength occurred. Furthermore, a similar standard deviation of 62.23 mm² and a mean difference of 187.42 mm² were observed between the control group and the experimental group. The results of the t-test showed that at a 95% confidence level, the t-value was −16.35 and the p-value was 0.00. This confirms that the incidence of bugholes was reduced in the experimental group where magnetic compaction was applied. This research may contribute to reducing the occurrence of bugholes in existing free-form concrete panels and securing both aesthetics and strength.

Meaningful public engagement is central to transportation planning, yet agencies face ongoing challenges in synthesizing large volumes of unstructured comments distributed across hearings, news media, and social platforms. Although natural language processing driven methods are increasingly adopted to analyze public engagement data, clear guidance remains lacking on which models are most suitable for specific data characteristics and analytical goals. This study compares transformer-based and classical natural language processing approaches for sentiment analysis and topic modeling in transportation infrastructure contexts. A curated multi source corpus from the North Houston Highway Improvement Project was developed, comprising Facebook posts, news articles, and public hearing documents. Sentiment classification using Bidirectional Encoder Representations from Transformers (BERT) models, specifically DistilBERT and RoBERTa, was benchmarked against lexicon-based approaches, while topic discovery using BERTopic was compared with classical probabilistic and matrix factorization topic models. Model performance was evaluated using classification accuracy and F1-Micro scores, topic coherence and interpretability, and consistency across platforms. Transformer-based methods outperformed classical approaches, particularly in informal or context rich settings where lexicon-based tools struggled with nuanced language, sentiment wavering, and multi-issue content. At the same time, BERTopic produced more coherent and transferable topic structures across heterogeneous engagement datasets, while lexicon-based models retained value for rapid screening. These results demonstrate that model selection should be guided by data characteristics and analytical objectives rather than reliance on a single technique. The study introduces a method selection framework that provides transportation agencies with practical guidance for analyzing diverse public engagement data.