Safety Analysis of Construction Work Zone Activities Using Convolutional Neural Network and BiLSTM

This study thoroughly explores the safety implications of construction activities on road safety, utilizing a robust data set of 1,500 daily progress reports from 2018 to 2020. We systematically categorize the reports into Low-, Mid-, and High-Safety Impact groups by employing state-of-the-art machine learning techniques, specifically Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN) classification models. Achieving an impressive overall accuracy of 78%, these models provide a comprehensive understanding of the safety landscape within the construction context. Furthermore, visually compelling word clouds for Mid- and High-Safety Impact reports unveil recurrent terms such as “ramp,” “truck,” “concrete,” “lane closure,” and “piping,” shedding light on specific construction activities correlated with elevated safety concerns. These findings contribute to a more profound comprehension of safety considerations in construction activities, potentially informing targeted interventions and strategies for enhanced road safety.