State-Based Analysis of General Aviation Loss of Control Accidents Using Historical Data and Pilots’ Perspectives

Neelakshi Majumdar, Purdue University


General Aviation (GA) encompasses all aircraft operations, excluding scheduled, military, and commercial operations. GA accidents comprise approximately 94% of all aviation accidents in the United States annually. 75% of these accidents involve pilot-related factors (pilot actions or conditions). Inflight loss of control means that the flight crew was unable to maintain control of the aircraft in flight. With almost 50% of loss of control accidents being fatal yearly, it continues to be the deadliest cause of GA accidents.The most common approach to understanding accident causation is analyzing historical data from sources such as the National Transportation Safety Board (NTSB) database. The NTSB database has abundant rich information. In contrast to the extensive investigations into and detailed reports on commercial aviation accidents, GA accident investigations tend to be shorter, and the resulting reports tend to be brief and limited—especially regarding human factors’ role in accidents. Only relying on historical data cannot provide a complete understanding of accident causation.There is a clear need to better understand the role of human factors involved in GA accidents to prevent such accidents and thus improve aviation safety. In my research, I focus on a specific type of accidents, inflight loss of control (LOC-I), the deadliest cause of GA accidents. I use historical data analysis and human-subjects research with pilots to investigate the role of human factors in loss of control accidents. Building on previous work, I created a state-based modeling framework that maximizes data extraction and insight formation from the NTSB accident reports by (1) developing a structured modeling language to represent accident causation in the form of states and triggers; (2) populating the language lexicon of states and triggers using insights from accident reports and pilots perspectives via surveys and interviews; and (3) applying Natural Language Processing (NLP) and machine learning techniques to automatically translate accident narratives into the language lexicon. The framework is focused on LOC-I but can be extended to other types of accidents. Findings from my study may help in consistent accident analysis, better accident reporting, and improving training methods and operating procedures for GA pilots.




Marais, Purdue University.

Subject Area

Artificial intelligence|Aerospace engineering|Occupational psychology

Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server