A construction collapse accident is one type of accident with high casualties and economic losses.Previous studies used different methods to investigate how causes affect the occurrence of collapse accidents.To explore the correlation and interdependence between causes of construction collapse accidents,231 construction collapse accident reports from 2015 to 2020 were collected.The R language platform was utilized to analyze the original accident reports for sampling field,segmenting words,deprecating words,and merging word groups.Then 43 collapse accident causative items were extracted from the original characteristic items,covering all direct and indirect causes in the construction collapse accident reports.A 0-1 reference matrix of 43 x231 dimensions was obtained according to the text mining results.Secondly,Python was employed to perform spectral clustering analysis on 43 causative items.The optimal number of clusters k=5 was determined by comparing the performance of the contour coefficient method and the elbow method on the data set.The Principal Component Analysis(PCA)was used to downscale the high-dimensional original reference matrix and performed spectral clustering based on the strength of association between causes.Finally,the association rule mining Apriori algorithm was adopted to determine the key causal combinations in the clusters.The association rules were measured by three metrics:support,confidence,and lift.The combinations with a strong correlation in the cluster were filtered out by setting a minimum confidence threshold and a minimum lift threshold of 1.0.The results indicate that 43 accident causative items are classified into 5 categories by spectral clustering,which is different from the traditional classification of accident causes based on human,machine,environment,and management.The association rules mining algorithm is used to get the key association combinations in each cluster,which reflect the correlation among the causes.Besides,the corresponding preventive measures are put forward for the five key cause combinations to reduce the construction collapse accident.
关键词
安全社会工程/建筑施工/坍塌事故/文本挖掘/谱聚类/Apriori算法
Key words
safety social engineering/construction/collapse accident/text mining/spectral clustering/Apriori algorithm
维普
万方数据