Minimally Supervised Topological Projections of Self-Organizing Maps for Phase of Flight Identification

TL;DR

This paper addresses phase-of-flight identification in general aviation using large-scale, per-second flight data that is sparsely labeled and highly imbalanced. It introduces minimally supervised self-organizing maps (MS-SOMs) that leverage topological projections and a graph-based nearest-labeled-neighbor consensus, enabling effective classification with only a small labeled subset. Empirical results show MS-SOMs can achieve accuracy comparable to a fully labeled naive SOM while using far fewer labels (about 30 per class) and exhibit improved robustness to class imbalance, particularly for minority phases. The work also provides a publicly available NGAFID-based dataset to support reproducibility and future development of data-efficient, real-world aviation safety analytics.

Abstract

Identifying phases of flight is important in the field of general aviation, as knowing which phase of flight data is collected from aircraft flight data recorders can aid in the more effective detection of safety or hazardous events. General aviation flight data for phase of flight identification is usually per-second data, comes on a large scale, and is class imbalanced. It is expensive to manually label the data and training classification models usually faces class imbalance problems. This work investigates the use of a novel method for minimally supervised self-organizing maps (MS-SOMs) which utilize nearest neighbor majority votes in the SOM U-matrix for class estimation. Results show that the proposed method can reach or exceed a naive SOM approach which utilized a full data file of labeled data, with only 30 labeled datapoints per class. Additionally, the minimally supervised SOM is significantly more robust to the class imbalance of the phase of flight data. These results highlight how little data is required for effective phase of flight identification.

Figures (4)

• Figure 1: Flow diagram for the data ingestion, selection, preprocessing, SOM training, and phase of flight estimation.
• Figure 2: An example topological projection of an unlabeled data point to its nearest labeled neighbors in a trained SOM topology.
• Figure 3: Different phases of a flight from its initial runway taxi phase to its final landing taxi phase.
• Figure 4: U-matrix for the best found 15x15 SOM trained with Flight 53438 data.