Perpustakaan Digital ITB

The air transportation industry's growth aligns with the number of safety reports submitted. With over 50% passenger increase in the past 16 years, accurately analyzing a high amount of safety reports is necessary. Narrative safety report data remain unexplored where hidden insights may lie inside those data. NASA ASRS data are an example of aviation safety narrative data publicly available with a voluntary reporting system. Exploring enormous amounts of data will be tedious if a human does it. It is necessary to use machine learning methods for exploring the narrative safety report data. Natural language processing can be utilized to extract text data using a computer. Unsupervised learning can be applied to text narrative data to automatically explore hidden insights useful for safety experts. This study aims to develop an explorative framework for analyzing the aviation safety narrative data to achieve a high-level view of the data. The framework is evaluated and demonstrated in a study case using NASA ASRS Boeing data from 2012-2021. Using Latent Dirichlet Allocation to perform topic modeling, the reports clustered into four main topics. The topics are then presented with a cloud-synchronized dashboard to give an upper-level view of the documents.