Analyzing Patterns of Multi-cause Accidents From KOSHA’s Construction Injury Case Reports Utilizing Text Mining Methodology

Construction accidents usually involve two or more injuries in succession considering various risk factors are present everywhere on site. This study aims to analyze the patterns of these multi-cause accidents through a text mining methodology. There were 1,300 accident reports from the Korea Occupational Safety & Health Agency (KOSHA). The collected data was refined and processed through a morpheme analyzer for semantic analysis. A Python algorithm was developed and applied to extract multi-cause accidents; 139 out of 987 accident cases were extracted. The occurrence patterns involving the 139 multi-cause accidents were based on the relationship of each accident type and the occurrence characteristics by type. The type of multi-cause accidents that occurred at the highest frequency were the narrowness or winded (Type 2) or fall (Type 1) due to the fall down or overturn (Type 5) of an object or structure. The rate of acting as a primary and secondary accident differed depending on the accident type. Falling (Type 1) and narrowness or winded (Type 2) had a very high proportion of secondary accidents, while the flying object, collision, fall down or overturn and collapse (Type 3, 4, 5 and 6, respectively) were more likely to act as primary accidents. Using the results from this study, once a specific accident is recognized, the scale of the accident can be minimized by closely examining the occurrence of similar accidents and possibly prevent future occurrences. Additionally, this study can provide direction to review data classified as a single accident from past instances.