Enhancing Wildlife Strike Risk Assessment through Large Language Model–Driven Data Analysis

Wildlife strikes remain a persistent safety and economic concern across global aviation operations, highlighting the need for advanced analytical methods to improve risk assessment and mitigation. Traditional statistical approaches to wildlife-strike data, while effective for structured variables such as altitude, phase of flight, or aircraft type, often overlook valuable insights embedded in the unstructured narrative components of strike reports. This study proposes the application of Large Language Models (LLMs) for the automated extraction, classification, and interpretation of information within the Federal Aviation Administration (FAA) Wildlife Strike Database. Using natural language processing (NLP) techniques, LLMs will be utilized to identify key entities (e.g., bird species, phase of flight, quarter of the year) and infer contributing factors to aircraft strike/damage cases. The resulting text-derived features will be integrated with structured data fields to support predictive modeling and pattern detection using machine learning algorithms. Expected findings include improved detection of latent relationships among variables such as strike height, weather conditions, and damage severity, along with greater predictive accuracy in assessing risk across different phases of flight. Beyond its analytical contributions, the study aims to explore how LLM-driven methodologies can be integrated into Safety Management Systems (SMS) to automate hazard identification, strengthen risk assessment processes, and provide decision-makers with interpretable, data-driven insights.