Enhancing Hazard Detection and Risk Severity Assessment in Construction through Multinomial Naive Bayes and Regression

Journal: Journal of Information Systems and Informatics

Publication: 2025-03-18 · Vol. 7 No. 1 · pp. 97–121

DOI: 10.51519/journalisi.v7i1.979

Abstract

This research delves into the crucial area of hazard detection and risk severity assessment within the construction industry, using machine learning techniques. The dataset utilized is from the Chinese Construction Company (CCECC), Uyo, Nigeria. Comprising over 100,000 instances, it captures various hazard categories prevalent in construction sites, providing a comprehensive foundation for predictive analysis. In the first phase of the study, the system is designed to detect hazards present in construction sites. Leveraging these data, the machine learning models are trained to predict potential hazards based on the information provided. Through TF-IDF vectorization, a feature extraction technique, the textual data is transformed into numerical representations. Multinomial Naive Bayes is employed for hazard classification due to its efficacy in handling text data, and with it, an accuracy of 0.99 was obtained. Subsequently, the trained model was evaluated to assess its performance and the severity of identified hazards are evaluated. The system quantifies the potential risk posed by each hazard using the risk severity attribute. Using the Linear Regression algorithm, the model predicts the severity of risks based on textual descriptions of a hazard.  In practical application, the research stresses the significance of risk management strategies in the construction industry to mitigate potential harm to personnel and infrastructure. This research contributes to advancing safety protocols within the construction sector, advocating for a culture of vigilance and precaution to address risks effectively.