2023 AGENDA

Machine learning is an increasingly mainstream technology that depends on large volumes of data. Aviation is a data driven environment. The intersection of Machine Learning and Aviation is a natural match. A quick review of ongoing research shows the use of Machine Learning across many facets of the aviation domain. There is research on data quality and anomaly detection, prediction of contrails, prediction of offshore precipitation, analysis of friction and breaking capability, and the modeling of aborted approaches (go-arounds) to name a few. The applications of Machine Learning in aviation will continue to grow and expand.

The aviation domain is a complex environment with a diverse set of stakeholders, organizations, and participants. There are overlapping policy domains, unknown or unclear data provenance, and data and knowledge representations that span decades. Adding to this complexity is the high dimensionality of aviation data sets. For example, a traffic flow management data set can contain hundreds for parameters. There are many multiple data sets across representations. Harmonization of operations and data sharing and collaboration are hallmarks of international ATM research.

In the machine learning domain, application frameworks often contain built in datasets. Classic examples of these datasets are things like the CIFAR-10, (Canadian Institute for Advanced Research) a labeled dataset of 60,000 samples, and the MNIST (Modified National Institute of Standards and Technology) dataset, a data set of the handwritten digits 0-9. These curated data sets are well tested and widely used. As the importance of data availability and quality has been recognized, modern frameworks now routinely ship dozens of these built-in datasets.

This work begins the process of identifying datasets for use in the aviation domain. The goal is to provide a standard data set to the aviation community to inform research and development.