A network intrusion detection system (NIDS) is a solution that mitigates the threat of attacks on a network. The success of a NIDS depends on the success of its algorithm and the performance of its method in recognizing attacks. We propose a deep learning intrusion detection system (IDS) using a pretraining approach with deep autoencoder (PTDAE) combined with a deep neural network (DNN). Models were developed using hyperparameter optimization procedures. This research provides an alternative solution to deep learning structure models through an automatic hyperparameter optimization process that combines grid search and random search techniques. The automated hyperparameter optimization process helps determine the value of hyperparameters and the best categorical hyperparameter configuration to improve detection performance. The proposed model was tested on the NSL-KDD, and CSE-CIC-ID2018 datasets. In the pretraining phase, we present the results of applying our technique to three feature extraction methods: deep autoencoder (DAE), autoencoder (AE), and stack autoencoder (SAE). The best results are obtained for the DAE method. These performance results also successfully outperform previous approaches in terms of performance metrics in multiclass classification.

网络入侵检测系统（NIDS）是一种缓解网络攻击威胁的解决方案。NIDS的成功取决于其算法的成功及其识别攻击的方法的性能。我们提出了一种使用深度自动编码器（PTDAE）和深度神经网络（DNN）结合的预训练方法的深度学习入侵检测系统（IDS）。使用超参数优化程序开发了模型。这项研究通过结合网格搜索和随机搜索技术的自动超参数优化过程为深度学习结构模型提供了替代解决方案。自动超参数优化过程可帮助确定超参数的值和最佳分类超参数配置，以提高检测性能。建议的模型已在NSL-KDD和CSE-CIC-ID2018数据集上进行了测试。在预训练阶段，我们介绍将我们的技术应用于三种特征提取方法的结果：深度自动编码器（DAE），自动编码器（AE）和堆栈自动编码器（SAE）。使用DAE方法可获得最佳结果。就多类分类中的性能指标而言，这些性能结果也成功地优于以前的方法。