Exploring Data Mining Representations of Track Data

Data mining is often used to analyze data that is too voluminous or complex to analyze by hand. However, most data mining algorithms require a fixed-length vector representation, in contrast to track data, which is naturally multi-dimensional and variable in length. We explore several methods for converting flight track data to a representation appropriate for data mining, and evaluate the performance of these representations in both clustering and classification tasks. Our results show that relevant features are captured in our representations, and describe the tradeoff in representational choices.