With the evolution of Internet of Things (IoT), there is a huge amount of data exchange among different smart devices located geographically apart from each other. As the data among these devices travels using an open channel, i.e., Internet, so one of the challenge is to build a secure system to protect from various cyber-threats. Cyber-criminals use new attack vectors to launch various attacks as the attack surface is also increasing exponentially. To deal with these threats, the network requires traffic surveillance along with strong access control policies. To handle the aforementioned issues, the existing proposals are found to be less efficient as most of these are not able to detect anomalies with change in the definition of attacks vectors. However, artificial intelligence and machine learning based techniques for network intrusion detection systems (NIDS) are found to be effective for such environment as they have revolutionized in the recent era for threat detection efficiently with high accuracy in a given time frame. Motivated from these facts, in this paper, we propose the taxonomy of existing machine learning based threat detection techniques for network intrusion detection. The existing techniques are compared with various existing solutions using different evaluation metrics. The comparative analysis demonstrate the applicability of one of the techniques with respect to its merits over the others. Lastly, we also discussed the open issues and challenges with future insights to the readers.

随着物联网（IoT）的发展，地理位置彼此不同的不同智能设备之间存在大量数据交换。由于这些设备之间的数据是通过开放通道（即Internet）传输的，因此面临的挑战之一就是建立一个安全的系统来防御各种网络威胁。随着攻击面也呈指数级增长，网络罪犯使用新的攻击媒介发起各种攻击。为了应对这些威胁，网络需要流量监控以及强大的访问控制策略。为了解决上述问题，发现现有建议的效率较低，因为其中大多数都无法检测到攻击向量的定义发生变化的异常情况。然而，人们发现，针对网络入侵检测系统（NIDS）的基于人工智能和机器学习的技术在这种环境下是有效的，因为它们在最近的时代已经发生了革命性变化，可以在给定的时间范围内高效，高效地进行威胁检测。基于这些事实，在本文中，我们提出了用于网络入侵检测的现有基于机器学习的威胁检测技术的分类法。使用不同的评估指标将现有技术与各种现有解决方案进行比较。比较分析证明了其中一种技术相对于其他技术的适用性。最后，我们还讨论了开放的问题和挑战，并为读者提供了未来的见解。