Air Safety Data Mining Research Ongoing

Air Safety Week, July 14, 2008

Computer scientists at The University of Texas at Dallas are developing technology that will sift through mountains of aviation data in search of ways to further enhance flight safety.

Part of a new three-year, $1 million NASA-funded project being done in collaboration with researchers at the University of Illinois at Urbana- Champaign, the work focuses on more than three decades of what are called "anomalous aviation events," or incidents that deviated from normal flight operations.

Using data-mining techniques that are increasingly popular in searching for kernels of relevant information within enormous amounts of data - crime statistics or genomics data, for instance - researchers at UT Dallas hope to identify subtle patterns of aviation events that could foreshadow future catastrophe.

The primary source for air safety information is the Aviation Safety Reporting System (ASRS) database.

UT Dallas researchers will be addressing two primary questions involving this aviation data: What anomalies are associated with a given aviation event, and why did the event occur. These issues are complicated, though, by the "noise" inherent in the data, which stems from typos and grammatical mistakes as well as the use of abbreviations and jargon not widely used outside the aviation industry.

"Non-aviation experts will find it difficult to understand these reports," said Dr. Vincent Ng, an assistant professor of computer science at UT Dallas and a co-principal investigator for the project. "Traditional text-mining techniques are unlikely to work well on this data because those techniques were developed primarily for use with clean text. So our biggest research challenge lies in developing techniques that can handle highly noisy data."

"It's essential to automatically mine large amounts of aviation safety reports in order to understand anomalous events and improve safety," said Dr. Latifur Khan, an associate professor of computer science and principal investigator for the UT Dallas portion of the research.

"With the rapid growth of the aviation industry worldwide and the increasing power and complexity of aircraft, spacecraft and other aviation systems, data related to aviation safety is growing rapidly, making large-scale manual exploration unrealistic. It's crucial to develop data-mining systems to analyze these aviation reports to help prevent future incidents," he added.

Air safety data-mining is not so easily done, though, due to the variety of such anomalous events - airspace violations, in-flight encounters with birds, miscommunication between pilots and flight controllers, to name a few - as well as the complexity of the aviation systems involved and the variety of people throughout the aviation industry who report the various incidents. The software required to accomplish the task must be sophisticated enough to discover patterns, correlations and trends within the jumble of information.

[Copyright 2006 Access Intelligence, LLC. All rights reserved.]