Delay Risk Detection in Road Construction Projects Utilizing Large Language Model

Construction projects worldwide consistently face the persistent challenges of time and cost overruns, wreaking havoc on budgets and economies. Effective mitigation necessitates identifying root causes—a complex task often dependent on deciphering project reports and subjective expert knowledge. Remarkably, there is a notable lack of suitable artificial intelligence (AI) techniques for addressing the complex challenges in road construction management. This study addresses this gap by employing a large language model (GPT-3), which is an advanced language model, to develop an unbiased delay risk detection framework. This research methodology involved a comprehensive literature review unveiling the limitations of conventional approaches that rely heavily on subjective data collection methods. To address these shortcomings, the proposed framework automates the generation of question–answer pairs using ChatGPT, thereby ensuring data consistency and efficiency. Real-time project reports serve as the crucible to validate the system’s efficacy, resulting in impressive performance metrics: high precision for “Class 0 (No Delay)” (0.79), “Class 1 (Delay Detected)” (0.91), and “Class 2 (False Positives)” (1.00), high recall for “Class 0” (1.00) and “Class 1” (0.89), a balanced F1-Score (0.90 overall), and an accuracy of 0.85, emphasizing its ability to minimize both false positives and false negatives. This study explored the potential of AI-driven solutions, such as GPT-3, to contribute to advancements in construction project management. Our framework presents a data-driven approach for forecasting project delays, enhancing monitoring capabilities, and bolstering sustainability.