Data Cleaning Process.
The purpose of data processing is to get clean, usable landing scenarios to be implemented into the TensorFlow Learn to Rank (TF-LTR) model. There was extensive data visualization and sorting done to obtain the finalized dataset. The following requirements were used to determine if an airplane fit into a usable landing scenario; geometric altitude of > 3000ft, landing at the Ames Airport (geometric altitude = 955ft), and more than one airplane landing within 7 minutes of another. All descent data of the airplane is kept for analysis, this includes all weather and ADS-B data on the airplane from 3000ft to 955ft in geometric altitude. Historical weather data was timestamp matched to the observed ADS-B data. For training purposes, historical ADS-B data was also used to compensate for the lack of usable landing scenarios collected from airport ADS-B data.
Figure. 1 Data Processing Flowchart
This is a flowchart that represents the data cleaning process, which shows how data went from collected at the Ames Airport to usable for TF-LTR modeling.
Figure 2. Data Visualization of Usable Landing Scenario
Data visualization was done to find usable landing scenarios. Airplane altitude data was plotted to visualize landing patterns of different airplanes, shown in colors. See an example of a usable landing scenario.