A hybrid Deep Learning Strategy for an Anomaly Based N-IDS

This paper presents a hybrid deep learning neural network for classifying the network traffic data. In this regard, a Stacked Autoencoder and Feedforward neural network with tangent activation function have been employed. Firstly, we pre-trained the stacked Autoencoder with unsupervised learning method to improve the generalization of the classifier and limit the over-fitting problem in building the Feedforward neural network. in this state, the data is reconstructed into a new representation. After that, the Feedforward neural network as a supervised classifier has been stacked on the top. The purpose was to map the data in this new representation into class predictions. Finally, we have fine-tuned the entire network to accomplish the optimal hybrid model. A k fold cross-validation has been conducted to validate the system. CICIDS2017 datasets has been used in the experiment to classify normal and abnormal behaviour. The experimental results obtained by analyzing the proposed system show their superiority in terms of accuracy, detection rate and false alarm rate as compared to two state-of the-art machine learning algorithms which are Support Vector Machine (SVM) and Deep Neural Network. Our study achieves 98%, 100% for accuracy rates and F1 score respectively.