Study on Machine Learning Methods for General Aviation Flight Phase Identification

Accurate identification of phases of flight is an essential step in analyses such as airport operation counts, fuel burn estimation, and safety studies. Past research has focused primarily on using positional data with rule-based or probabilistic-based decision-making to identify the phases of flight. Many of these efforts note that the task of correctly identifying phases of flight is challenging, often requiring extreme fine-tuning of methods. In this paper, we initially study whether combinations of dimensionality reduction of flight data records from general aviation aircraft impact clustering into the correct flight phases (climb, cruise, or descent) without any preprocessing or fine-tuning. For dimensionality reduction, we considered the low variance filter, the high correlation filter, principal component analysis, and autoencoders. We found that these dimensionality reduction algorithms do not offer any benefit for the phase identification task, as compared to feature selection that simply omits engine-specific features. For the clustering task, we considered K-means and Gaussian mixture models. After performing clustering on eight test flights, we conclude that both methods are adequate at identifying the phases of flight in various general aviation flights and yield similar results.