QAR (quick access recorder) data exceedance and statistical analyses is used to flight safety event reasoning. Machine learning algorithms are applied to fully automatically classify pilot behaviors into classes. Both classifications (supervised) and clustering (non-supervised) methods are tried, for instance, gentle Adaboost (Adaptive Boosting) classification, K-means and Gustafson-Kessel (G-K) clustering. Routine exceedance events are assigned to gentle adaboost labels for training model and evaluating testing result, while K-means and G-K clustering are conducted in free-style. All results are compared, and findout unveils that each method emphasizes different series of pilot performances given by a method named Minimum Redundancy Maximum Relevance (mRMR). After theoretical study, we realize in our cases, G-K clustering is much more accurate than K-means in clustering methods.
Application of Machine Learning in Flight Safety Event Reasoning
2021-10-20
1314190 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Reasoning-Based Framework for Driving Safety Monitoring Using Driving Event Recognition
Online Contents | 2013
|MACHINE LEARNING ENABLED TURBULENCE PREDICTION USING FLIGHT DATA FOR SAFETY ANALYSIS
British Library Conference Proceedings | 2021
|Flight safety event management method as illustrated by a guided reactivity model
DOAJ | 2021
|