achevrot / scifly Goto Github PK

Hello. Could your model detect anomalies of flights that do not exist in the training set? For example, European flights for training, while Asian for testing? Thank you.

The ADS-B features involved in the training include Tracking Delta. Are you the first person to propose this feature for flight anomaly detection? Why choose it as one of the features? Is it to solve the problem of cyclic features? Is it based on the assumption that most flights fly in a straight line from the departure to the arrival airport? Does the messages received by the ADS-B receiver include the location of the arrival airport, or can it be obtained from other channels?

When I tried to run the file "data_cleaning.py", I found that the configuration file (configFilePath=r'../config/data_cleaning.ini') does not seem to exist, which leads to the error of "configparser.NoSectionError: No section: 'paths'". Similarly, other py files seem to have the same problem. How could I deal with it?

As mentioned in Section 4.1, some external contextual factors may alter the shape of the time windows over time, hence every time window is associated with a contextual feature. I think this may mean that the three different flight phases have different window sizes. However, in Fig.2 (architecture of CAE model), the window size is fixed. Could you explain in detail how to set the window size in the CAE model?

I encountered some unexpected errors when reproducing your experiment. Could you share the Python and Tensorflow versions used in your experiment? Does your experiment depend on some specific version of other modules?