MINDFUL is an Deep Learning model that performs a binary classification by combining autoencoders and 1D Convolutional Neural Network.

Read more: Multi-Channel Deep Feature Learning for Intrusion Detection

This repo contains the classifier only. For the full network intrusion detection system from the paper and all experiments, please see https://github.com/gsndr/MINDFUL.

class MindfulNET.MINDFUL.MIDFUL_NET(dsConfig, autoencoderA=None,autoencoderN=None,model=None)

• dsConfig: a dictionary of parameters dsConf = {'pathModels': 'models/AAGM17/', 'testName': 'AAGM17'} that indicate the path and the name of saved files and models
• autoencoderA: path of learned autoencoder fo class with label 0
• autoencoderN: path of learned autoencoder fo class with label 1
• model: path of learned 1DCNN

The default path of Deep Learning models is None. If the path of a model is setted the framework load the learned model from path otherwise the framework find the better model by performing a

fit(X, y)

Fit the model according to the given training data

• X: array of shape (n_samples, n_features)
• Y: target array of shape (n_samples,)

• autoencoderA: learned autoencoder for class with label 0
• autoencoderN: learned autoencoder for class with label 1
• autoencoderN: learned 1DCNN

predict(X)

Predict class for X.

• X: array of shape (n_samples, n_features)

• y_pred: array-like of predicted values shape (n_samples,)

predict_proba(X)

Predict probabilities for samples in X.

• X: array of shape (n_samples, n_features)

• y_prob: array of the class probabilities of the input samples of shape (n_samples, n_classes)

The code relies on the following python3.6+ libs.

Packages need are:

• Tensorflow
• Keras 2.3
• Pandas 0.23.4
• Numpy 1.15.4
• Scikit-learn
• Hyperopt
• Hyperas

Here is a simple example of how to make a MINDFUL object:

    sys.path.insert(1, 'MindfulNET')
    from MindfulNET.MINDFUL import MINDFUL_NET
    
    dsConf = {'pathModels': 'models/AAGM17/', 'testName': 'AAGM17'}
    clf=MINDFUL_NET(dsConf,autoencoderA=pathModels + 'autoencoderAttacks.h5',)
    clf.fit(X,Y)
    Y_pred = clf.predict(X_test)

    cm = confusion_matrix(Y_test, Y_pred)
    print('Prediction Test')
    print(cm)

If you pass the path to the learned model the code kip the pearning phase of the model (e.g. in the sample above the autoencoder on class 0.

As a quick start, a demo script is provided in example.py. You can either run it directly or enter the following into your python console python3 example.py -d AAGM -i MINDFUL.conf

In the file Example_Prediction.py a sample of MINDFUL used only for prediction. You can pass on MINDFUL class the path of the models to skip the learning phase.

Please cite our work if you find it useful for your research and work.

  @ARTICLE{9036935, 
  author={G. {Andresini} and A. {Appice} and N. D. {Mauro} and C. {Loglisci} and D. {Malerba}}, 
  journal={IEEE Access}, 
  title={Multi-Channel Deep Feature Learning for Intrusion Detection}, 
  year={2020}, 
  volume={8}, 
  number={}, 
  pages={53346-53359},}