Predicting fatal fall from heights accidents using random forest classification machine learning model

• Main Authors: Zermane, Abderrahim, Mohd Tohir, Mohd Zahirasri, Zermane, Hanane, Baharudin, Mohd Rafee, Mohamed Yusoff, Hamdan
• Format: Article
• Published: Elsevier 2023
• Online Access: http://psasir.upm.edu.my/id/eprint/110072/; https://www.sciencedirect.com/science/article/pii/S0925753522003629
• Institution: Universiti Putra Malaysia

The focus of this paper is to use machine learning to create a prediction model that detects the probable factors impacting fatal falls from heights accidents at the Malaysia construction industry. The dataset used in this study was imported from the Department of Occupational Safety and Health of Malaysia’s industrial accident database. The dataset details 3321 accident scenarios which include; the date, the activity, the region, the summary of the accident, the direct cause, the root cause, and the factors. Seven machine learning models were tested to determine which model fits the dataset better, and as a result, the Random Forest Classification model was selected for this work. Random Forest classification tested several contributing factors such as: site conditions, management factors, individual characteristics and agent factors separately to determine their accuracy. Management factors and individual characteristics factors recorded the highest accuracy in every other prediction model; while agent factors recorded the highest accuracy in the random forest model. Additionally, this approach created ensemble predictions based on all of the dataset's characteristics. As a result, this study establishes the feasibility of machine learning in the field of construction safety management. The provided results can aid in accident prevention by increasing awareness of potential safety hazards, quantitatively predicting fatal accidents, and implementing the findings in potential safety management systems.