Page 10 - IJAMD-2-1

International Journal of AI for
Materials and Design
Utilizing AI for NTSB UAV accident categorization

and rotor malfunctions. Figure 1 shows the number of dataset with assigned CICTT categories could create
accidents per category. opportunities for estimating predictive precision of the
Geographic distribution analysis showed that UAV proposed models.
accidents are widespread across the United States, with 5. Discussion
certain areas exhibiting higher concentrations of incidents.
Hotspots were identified in regions with dense urban The results of this study underscore the critical importance
development and higher UAV activity. Figure 2 shows the of addressing system and component failures to enhance
map of the incidents. UAV safety. The prevalence of issues such as loss of control
and navigation system failures suggests that technological
This spatial analysis is crucial for targeted safety improvements and stringent maintenance protocols are
interventions and regulatory measures in specific areas. essential. The geographic distribution of accidents further
Visualizations, including heat maps and cluster maps highlights the need for localized safety interventions,
created using Folium, effectively illustrate these geographic particularly in urban areas where UAV operations are more
patterns, providing clear insights into regional accident frequent and complex. Monitoring the annual evolution
trends. can provide valuable insights for anticipating specific
Figure 3 shows notable yearly variations, with a risk trends, allowing for more effective preemptive safety
significant spike in accidents observed in 2019. However, measures.
a downward trend in accidents per UAV after 2019 When compared with previous studies, our findings
suggests that recent safety measures and technological align with the broader consensus that UAV safety is
advancements are beginning to have a positive impact. predominantly compromised by technological failures.

The visualizations and tables generated from the data However, our use of AI-driven categorization offers a
2
analysis provided a comprehensive view of the accident more nuanced understanding of these issues. Ferrigan
data. For instance, a table summarizing the frequency of identified similar hazards but his findings lacked the
different accident causes offered a quick reference to the granularity provided by AI categorization. In addition,
most common issues, while a series of bar charts illustrated our identification of specific geographic and temporal
the monthly and yearly accident trends. Interactive maps patterns offers new dimensions for understanding UAV
highlighted accident hotspots, enabling a more intuitive safety, which were less explored in prior research. This
understanding of the geographic distribution. These visual highlights the added value of using advanced AI and data
tools not only supported the findings but also enhanced visualization techniques.
the overall presentation of the data, making it accessible The integration of AI and data visualization has
and interpretable for a broader audience. significant implications for improving UAV safety
Some limitations associated with the work are worth policies. AI, particularly NLP through GPT-4, enables
mentioning. A significant one is the limitation of the efficient and accurate categorization of accident reports,
number of reports that supported the analysis. Expanding which is essential for large-scale data analysis. This
this initial exploratory approach to a larger set of reports automation reduces the manual workload and increases
would allow the testing of model predictive performance the consistency of data interpretation. Data visualization
with added confidence. Developing an SME validated tools like Matplotlib, Seaborn, and Folium transform raw
data into insightful visual representations, making it easier
for policymakers to identify trends and patterns. Together,
these technologies provide a powerful framework for
developing data-driven safety strategies, enhancing
regulatory measures, and ultimately reducing the incidence
of UAV accidents.
6. Future work

Future research should explore the potential of AI techniques
for more sophisticated analysis of UAV accident data. These
techniques can uncover complex patterns and correlations
simpler models might miss. Further development of NLP
methods is crucial for extracting deeper insights from
Figure 1. Number of accidents in each category accident reports. Advanced NLP models can analyze

Volume 2 Issue 1 (2025) 4 doi: 10.36922/ijamd.8544