Safety assessment in megaprojects using artificial intelligence

•LCCA improved the prediction performance of AI methods with heterogenic data.•ANN and CBR can deal with the dataset including plentiful textual expression as incident cases.•CBR showed better performance over ANN in the prediction process.•CBR is more adaptable to new case entries than ANN for occu...

Full description

Saved in:

Bibliographic Details
Published in:	Safety science 2019-10, Vol.118, p.273-287
Main Authors:	Ayhan, Bilal Umut, Tokdemir, Onur Behzat
Format:	Article
Language:	eng
Subjects:	Artificial intelligence Artificial neural networks Case-based reasoning Cluster analysis Clustering Construction industry Construction safety Data analysis Domains Heterogeneity Latent class clustering analysis Megaprojects Neural networks Occupational safety Prediction models Safety Safety management Subgroups
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	•LCCA improved the prediction performance of AI methods with heterogenic data.•ANN and CBR can deal with the dataset including plentiful textual expression as incident cases.•CBR showed better performance over ANN in the prediction process.•CBR is more adaptable to new case entries than ANN for occupational safety data.•Record keeping is the key element of preventative actions for megaprojects.•The use of AI-based systems can prevent incidents in megaprojects. Record keeping systems are gaining more importance in addressing safety problems in megaprojects. The information recorded is being converted into large data domains since it has become necessary to examine problems down to the last detail to deal with them properly. Due to a high number of attributes and the type of information, the data has a high level of heterogeneity. The aim of this paper is to propose an innovative safety assessment methodology to predict the possible scenarios and determine preventative actions. The study introduces predictive models based on factual information and shows how to deal with the safety data which has a significant level of heterogeneity. Latent Class Clustering Analysis (LCCA) was performed to reduce the heterogeneity and extract homogenous subgroups from the data. For predictive purposes, Artificial Neural Network (ANN) and Case-Based Reasoning (CBR) were employed to estimate the outcome of incidents in terms of severity. CBR gave promising results and exhibited better performance than ANN. CBR achieved results in 86% of the test cases with 18% error at most. Moreover, the overall prediction accuracy of fatal incidents was equal to 86.33%. As a final step, the study presented preventative actions to eliminate safety failures. Ultimately, the proposed methodology can assist construction industry professionals in examining future safety problems by utilizing the collected large scale data. Furthermore, the study provides necessary preventative measures to be implemented before and during the construction stage.
ISSN:	0925-7535 1879-1042