With the emergence of ever-advancing network threats, the guarantee of system security becomes increasingly crucial, especially in the dynamic and decentralized ad-hoc networks. One essential part of cybersecurity is intrusion detection, which identifies anomalous activities according to traffic patterns. However, the class-imbalanced data have caused a challenging problem where the number of abnormal samples is significantly lower than that of the normal ones. This class imbalance problem confines the performance of intrusion classifiers and results in low robustness to unknown anomalies. In this paper, we propose a novel Imbalanced Generative Adversarial Network (IGAN) to tackle the class imbalance problem. In the primary novelty of our model, we introduce an imbalanced data filter and convolutional layers to the typical GAN, generating new representative instances for minority classes. Further, an IGAN-based Intrusion Detection System, namely IGAN-IDS, is established to cope with class-imbalanced intrusion detection, using the instances generated by IGAN. Concretely, IGAN-IDS consists of three modules: feature extraction, IGAN, and deep neural network. First, we utilize a feed-forward neural network (FNN) to transform raw network attributes into feature vectors. Then, the IGAN generates new samples expressed in the latent space. Finally, the deep neural network, composed of convolutional layers and fully-connected layers, executes the final intrusion detection. We conduct experiments on three benchmark datasets to evaluate the performance of IGAN-IDS, comparing against 15 other methods. The experimental results demonstrate that our proposed IGAN-IDS outperforms the state-of-the-art approaches.

随着不断发展的网络威胁的出现，对系统安全的保证变得越来越重要，尤其是在动态分散的ad-hoc网络中。网络安全的重要组成部分是入侵检测，它可以根据流量模式识别异常活动。但是，类别不平衡的数据引起了一个挑战性的问题，其中异常样本的数量明显低于正常样本的数量。该类不平衡问题限制了入侵分类器的性能，并导致对未知异常的鲁棒性较低。在本文中，我们提出了一种新颖的不平衡生成对抗网络（IGAN）来解决阶级不平衡问题。在我们模型的主要新颖之处中，我们为典型的GAN引入了不平衡数据过滤器和卷积层，为少数群体产生新的代表性实例。此外，使用IGAN生成的实例，建立了一个基于IGAN的入侵检测系统，即IGAN-IDS，以应对类不平衡入侵检测。具体来说，IGAN-IDS包含三个模块：特征提取，IGAN和深度神经网络。首先，我们利用前馈神经网络（FNN）将原始网络属性转换为特征向量。然后，IGAN生成在潜在空间中表达的新样本。最终，由卷积层和全连接层组成的深度神经网络执行最终的入侵检测。我们对比了其他15种方法，对三个基准数据集进行了实验，以评估IGAN-IDS的性能。