Predicting Accident Outcomes in Cross-Border Pipeline Construction Projects Using Machine Learning Algorithms

Abstract

Pipeline construction projects are necessary to provide gas and liquid energy transportation. Although various studies have investigated the contributing factors to accidents occurring in pipeline construction projects, there is a need for a predictive model for such incidents. The main purpose of this study is to provide an artificial intelligence-based model to predict the outcomes of occupational accidents. In this context, 1184 incident cases, including injury, near-miss, and asset-product damage, taken from a pipeline construction project constituted the primary dataset. Twelve prediction models are formed by changing the input domain according to the type of incident, time, and cause type attributes, leading to 12 distinct sub-datasets. Then, each dataset is tested with 11 different machine learning (ML) algorithms to derive an effective prescriptive model. The descriptive results show that low awareness of job hazards and improper vehicle operations were the most critical immediate causes, while failure in risk recognition and site supervision were the major root causes of pipeline construction accidents. Among the ML methods, the deep learning algorithm performed better than its counterparts in eight sub-datasets. Finally, a prescriptive model incorporating the ML application procedure is recommended for construction companies to reduce occupational accidents. Overall, the proposed model and findings are expected to contribute to preventing and reducing construction accidents in pipeline projects by adopting relevant strategies.