Commercial Aircraft Flight Data for Heterogeneous Anomaly Detection

Several hundred flight parameters are recorded per second from worldwide commercial aircraft operations, including guidance, navigation, control, avionics, propulsion, and pilot inputs. This dataset is associated with a novel multiple kernel learning algorithm designed to detect safety anomalies in mixed discrete and continuous data streams. The algorithm was tested on real-world aviation data and reportedly uncovered operationally significant events not detectable by other methods.

Use Cases

• Developing anomaly detection models based on mixed discrete and continuous flight parameters.
• Researching causal influences between discrete event sequences and continuous system performance.
• Benchmarking new machine learning methods against state-of-the-art aviation safety analytics.
• Proactive safety monitoring to identify atypical sequences of events in flight operations.

Strengths

• Data is recorded at a high frequency of one-second intervals for the duration of flights.
• Includes heterogeneous data types: continuous measurements as well as binary and categorical measurements.
• Derived from a real-world, complex system with worldwide commercial fleet operations.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• License is closed, restricting potential use cases.

Provenance

Source

paperswithcode

Collection Method

Likely collected from commercial aircraft flight data recorders.

Geography

Worldwide commercial aviation operations.