Using LSTM to detect spoofing attacks in an Air-Ground network

Specifications

• IDE: Spider
• Packages:
  • Tensorflow 2.1.0
  • Keras
  • NumPy
  • Scikit-learn
  • Matplotlib

Datasets:

• Training dataset is trainX_H0__LSTM_IN.npy that contains normal time-series data samples.
• Testing datasets include testX_H0__LSTM_IN.npy and testX_H1__LSTM_IN.npy.
  • testX_H0__LSTM_IN.npy contains normal time-series data samples
  • testX_H1__LSTM_IN.npy contains abnormal time-series data samples.
  • Note that H0 means the hypothesis that there is no spoofing attack, and H1 means the hypothesis that there is a spoofing attack from some spoofers/impersonators.
• The shape of each dataset is (82, 15, 15), where the first number (82) is the length of a time-series data sample, the second number (15) is the number of previous time slots that we want to look back for learning-from-the-past purposes, and the last number (15) is the number of receive antennas (or the number of features).
• The following figure illustrates 2 time-series data samples, corresponding to H0 and H1, respectively.

Goals

• Train an LSTM autoencoder in order for it to learn the H0 data samples.
• Once the LSTM autoencoder has been trained, it can capture the most significant characteristics of the H0-normal data.
• Test whether the testing datasets contain H1-abnormal data samples that are associated with spoofing attacks.
• A detection rule relies on contrasting the output and the input of the LSTM autoencoder. Imagine that if the input is a H0-normal data sample, then the output should look similar to the input. In this case, the difference between the input and the output is insignificant. On the other hand, if the input is a H1-abnormal data sample, then there is a big difference between the input and the output, because the trained LSTM autoencoder is meant to learn normal data samples.

Results

• Some results are stored in the folder saved_figs.