Blender-GAN: Multi-Target Conditional Generative Adversarial Network for Novel Class Synthetic Data Generation

With the rise in usage of computer networks across the globe, large amounts of data are being generated which has raised the need for robust intrusion detection systems to monitor the network traffic. Organizations are leveraging the predictive power of machine learning and deep learning models to develop robust IDS which can alarm any illicit intrusion. However, the power of deep neural networks is often constrained by the availability of training data. In real-world scenarios where the acquisition of large training datasets is not possible, synthetic data generation is emerging as a promising alternative. Conditional GAN allows us to condition the network with additional information such as class labels and generate data resembling the input class. The limitation of this architecture is that it works with a single label and doesn't allow mixing of labels to create newer varieties of data. We propose Blender-GAN, a generative adversarial network, that takes multiple class labels and their respective proportions to generate a novel output that represents all input classes in the input proportions. Unlike traditional GANs two discriminators are used, one traditional Real/Fake classifier and another to control the class of generated data. During the new data generation phase, multiple classes are blended in defined proportions which creates new data with characteristics of the classes in the specified ratio. The proposed architecture was implemented for generating synthetic network intrusion data, using various proportions of different known attack classes. The Blender-GAN output generated new attack classes, which resembled realistic network data.