Researchers Develop Techniques to Analyze Flight Safety Data

5/12/2009

Written by

Jiawei Han
Jiawei Han

Aided by a $1.1 million grant from NASA, computer science researchers at the University of Illinois are investigating new data mining techniques for analyzing flight safety data. The project aims to find computing methods to better understand anomalous flight events and improve airline safety.

The work focuses on more than three decades of what are called "anomalous aviation events," or incidents that deviated from normal flight operations. Data relating to more than 150,000 such events are stored in a system known as the Aviation Safety Reporting System (ASRS). The variety of such events - airspace violations, in-flight encounters with birds, and miscommunication between pilots and flight controllers, to name a few - the complexity of the aviation systems involved, the heterogeneity of data sources, and the variety of people who report such events all contribute to the difficulty of discovering trends and correlations in the data.

Computer science professor and Information Trust Institute (ITI) member Jiawei Han, an expert in data mining, data warehousing, spatiotemporal data analysis, and stream data mining, is leading the effort to develop new tools and algorithms to make sense of this jumble of information. The research team also includes Illinois CS professor Cheng Zhai, an expert in information retrieval, natural language processing, machine learning, and text mining, and researchers from UT-Dallas and Boeing Corporation.

"Since the root causes of such events can be complicated and the reporters' opinions may vary or are inconsistent, we believe that an effective way to help analysts is to provide a software environment that can be used to interactively mine the reports to obtain interesting patterns," says Han. "We want to create a system that enables users to flexibly navigate through the event and pattern space to understand the linkages among different events and patterns."

Ultimately, says Han, the team wants to enable analysts to go beyond the raw text information in the reports to interact more directly with the knowledge buried in the reports. To reach this goal, the team will employ data mining and contextual text mining techniques, among others, to classify, cluster, and build links between data elements.

Researchers hope that their system enables another crucial task beyond the detection of anomalous events: the prevention of similar incidents in the future.

"Our intent is to develop a system that will lead to the discovery of patterns, correlations, trends, and causality in the data so that flight safety analysts can have a better understanding of incidents and the contextual factors that may have influenced their occurrences," says Han.

The new collaboration emerged in part out of a project Han leads in ITI's Boeing Trusted Software Center, addressing online mining of anomalous moving objects for security protection. Boeing researchers were excited about that work and interested in pursuing additional research with him. For the new project, they are contributing the expertise of one Boeing researcher.


Share this story

This story was published May 12, 2009.