Classification of Aviation Safety Reports using Machine Learning

The growing aviation industry in both number of passengers and aircraft movements in combination with the implementation of Just Culture facilitates a growing number of occurrence reports being submitted by service providers to their competent authority. Still, actually using the sheer amount of safety data being collected with these reports for safety initiatives has proven to be difficult as analysing the reports is a time consuming activity. Moreover, reports usually suffer from missing data and labelling. This paper demonstrates an automatic text classification method by means of machine learning. By automatically classifying occurrence reports analysts may be able to work more efficiently as conducting queries on reporting databases become more accurate thanks to the categorization of reports. In this paper, a text classifier is trained by means of a Random Forest algorithm on the ICAO Occurrence Category. The classifier is used to classify more than 45,000 reports while reaching accuracy between the 80–93 %.