Intrusion Detection System Using Recurrent Neural Networks and Attention Mechanism
Praveen Kumar Kollu · International Journal of Emerging Trends in Engineering Research · 2019
Protecting sensitive information over the internet requires specialized algorithms that can detect subtle abnormalities in the data.These abnormalities are harder to detect as most of the data contains redundant information that does not contribute to classification of normal and abnormal classes.Finding the most optimal features is still a difficult task.To optimize the model to focus on the most relevant features and to make the model train considerably faster, we are proposing an attention based recurrent neural network for intrusion detection in large scale networks.The model is optimized to leverage the power of GRU network while focusing on the most important features for classification.The latter is achieved by the attention vector that is introduced to the network.The efficiency of the network is determined using several measures such as accuracy, precision, recall and F1-score.These measures are compared among state-of-the-art classification algorithms to determine how our proposed model performs on par with the current approaches.