Abstract
Here, we present a robust method for detecting flight automatically for use in digital luggage tags. The method is based on simple statistical aggregates using air pressure and 3D accelerometer measurements and complies with IATA and FAA requirements and recommendations. We achieve a correct detection of flight phases for 98.4% of the recorded data samples (with an ROC AUC score of 98.8%).
Data was recorded onboard several commercial flights in Norway, both in cabin and in the luggage compartment. Due to technical setbacks during the project period and the COVID-19 pandemic severely reducing the number of flights, the data available for analysis and method development was very limited. To mitigate the risk for severe overfitting, we used several countermeasures such as diligent cross-validation and the choice to keep model complexity low.
Data was recorded onboard several commercial flights in Norway, both in cabin and in the luggage compartment. Due to technical setbacks during the project period and the COVID-19 pandemic severely reducing the number of flights, the data available for analysis and method development was very limited. To mitigate the risk for severe overfitting, we used several countermeasures such as diligent cross-validation and the choice to keep model complexity low.