With the increased dependence on the internet for day to day activities, the need to keep the networks secure has become more vital. The quest of securing the computer systems and networks, from the users with destructive mindset, has resulted in the invention of surfeit devices and methods. One such method against whom the responsibility of discriminating between normal and harmful data, flowing on the network is, intrusion detection system (IDS). In this work an IDS model based on support vector machines is proposed. In order to enhance the detection capability of support vector machine based model for intrusion detection, and to eliminate the inherent problem of intrusion detection i.e, low accuracy of the system in detecting user to root and remote to local attacks, this paper proposes to use recent non-linear dimension reduction technique to enhance the discrimination of the data. Results demonstrate that t-SNE based dimension reduction improve the accuracy of SVM for network intrusion detection system. A comparison of the proposed system with the previous works has proven that this work has enhanced detection rate for almost all the attack groups.

中文翻译：

---

基于t-SNE的非线性降维用于网络入侵检测

随着日常活动对互联网的日益依赖，保持网络安全的需求变得越来越重要。对具有破坏性思维方式的用户确保计算机系统和网络安全的追求导致了过多设备和方法的发明。入侵检测系统（IDS）是一种可以区分网络上正常数据和有害数据的方法。在这项工作中，提出了一种基于支持向量机的IDS模型。为了增强基于支持向量机的入侵检测模型的检测能力，并消除入侵检测的内在问题，即系统检测用户到本地和远程的本地攻击的准确性较低，本文提出使用最近的非线性降维技术来增强数据的判别力。结果表明，基于t-SNE的降维技术提高了SVM在网络入侵检测系统中的准确性。将建议的系统与以前的工作进行比较，证明该工作提高了几乎所有攻击组的检测率。