Chemical Incident Data Mining and Application to Chemical Safety Analysis

摘要

The available US chemical incident databases present problems such as overlaps, missing data, and incomplete and inaccurate information as well as variations in definitions, scope of chemicals, and threshold quantities. When these data are used for analysis and developing accident trends, the uncertainties in the raw data propagate to the results and thus the conclusions drawn from the analyses also become suspect. This paper provides a definition of "incident," with examples of included and excluded events meant to clarify the scope of the definition, and describes how meaningful chemical incident information can be mined from existing public access databases. The data mining process consists of a detailed database merging protocol; which includes sanitizing data, eliminating duplicate records and overlaps, and identifying and eliminating events that do not meet the incident definition criteria. The analyses of these mined data can be used to develop meaningful trends, and this information can be used to measure the chemical safety performance. By comparison of the mined data with other sources it is possible to get a picture about the completeness and accuracy of our knowledge about chemical safety. Conclusions and assessment about what should be done to improve the chemical incident reporting system are derived from this work.