Identification of Construction Safety Risks Based on Text Mining and LIBSVM Method

The construction industry is an important pillar of the national economy in many countries, but it is also regarded as one of the most dangerous industries. To lower the rate of construction accidents and facilitate national economic growth, safety control is always the most frequently discussed topic in recent years. In China, government supervision plays an important role in this part by taking regularly inspections on construction sites to ensure control of construction safety. For safety risks such as improper operations or unsafe facilities, the government will take an on-site record and then send formal rectification notices to the contractor to improve the safety condition in time. Based on the methods of text mining and machine learning, this paper analyzes these formal notices issued by the Shanghai Municipal Construction Engineering Safety Quality Supervision Center during 2017–2018. Through text mining, 4,801 rectification notices are first processed by word segmentation, followed with word frequency statistics to obtain safety hazard events with high frequency. Then a library for support vector machines (LIBSVM), integrated software for support vector classification, is used to further classify these potential risks. The result has proved the possibility of the LIBSVM-based construction safety factors classification model for both construction contractors and supervision authorities to better control the overall construction safety situation. The study also reveals that facilities like scaffolds, tower crane, and foundation pit may easily turn into potential hazard sources if they cannot meet corresponding regulations. The findings contribute to a deepened understanding of indicators and facilities affecting construction safety which presents a high frequency of occurrence in construction processes.