Accident Analysis for Construction Safety Using Latent Class Clustering and Artificial Neural Networks

Ayhan B. U., Tokdemir O. B.

JOURNAL OF CONSTRUCTION ENGINEERING AND MANAGEMENT, vol.146, 2020 (SCI-Expanded) identifier identifier identifier

  • Publication Type: Article / Article
  • Volume: 146
  • Publication Date: 2020
  • Doi Number: 10.1061/(asce)co.1943-7862.0001762
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, Aerospace Database, Applied Science & Technology Source, Business Source Elite, Business Source Premier, Communication Abstracts, Compendex, Computer & Applied Sciences, ICONDA Bibliographic, INSPEC, Metadex, Public Affairs Index, DIALNET, Civil Engineering Abstracts
  • Keywords: Occupational health and safety in construction, Artificial neural networks (ANN), Latent class clustering analysis (LCCA), IDENTIFYING ROOT CAUSES, MODELS, PREVENTION, MANAGEMENT, RISK
  • Middle East Technical University Affiliated: Yes


Despite many improvements in safety management, the construction industry still has the highest potential for occupational injuries including High Severe (HS) work events, which result in injuries or fatalities, and Low Severe (LS) work events, which cause near misses or nonserious injuries. The analysis of incidents is highly dependent on the quality of records. Problems in recording and the heterogeneity of incident data may create conflicts while analyzing the relationship between attributes. The objective of the study was to develop a novel model to predict the outcomes of construction incidents using Latent Class Clustering Analysis (LCCA) and Artificial Neural Networks (ANNs) and determine necessary preventative actions. ANN has been used for many years to investigate the nonlinear relation between attributes and generate a logic between them. Herein, ANN was used to perform severity analyses of incidents utilizing real data, which were collected from various construction sites anonymously. Many factors affect the performance of ANN, including the size of the input and the heterogeneity of data. LCCA was used to seek out better performance and accuracy in ANN applications by reducing the heterogeneity of the incidents. By applying LCCA, attributes that possess different probabilities were clustered together and put into the ANN model. Then, the study concluded by providing a necessary preventative measure according to the result of incidents forecasted in advance. The research has two significant contributions. First, the hybrid model revealed promising results as the performance of the ANN-based predictive model was enhanced by addressing the heterogeneity of data. Second, the study presented professionals with practical preventative actions to avoid construction incidents according to the results of prediction.