Intrusion detection systems (IDSs) built on artificial intelligence (AI) are presented as latent mechanisms for actively detecting fresh attacks over a complex network. Although review papers are used the systematic review or simple methods to analyse and criticize the anomaly NIDS works, the current review uses a traditional way as a quantitative description to find current gaps by synthesizing and summarizing the data comparison without considering algorithms performance. This paper presents a systematic and meta-analysis study of AI for network intrusion detection systems (NIDS) focusing on deep learning (DL) and machine learning (ML) approaches in network security. Deep learning algorithms are explained in their structure, and data intrusion network is justified based on an infrastructure of networks and attack types. By conducting a meta-analysis and debating the validation of the DL and ML approach by effectiveness, used dataset, detected attacks, classification task, and time complexity, we offer a thorough benchmarking assessment of the current NIDS-based publications-based systematic approach. The proposed method is considered reviewing works for the anomaly-based network intrusion detection system (anomaly-NIDS) models. Furthermore, the effectiveness of proposed algorithms and selected datasets are discussed for the recent direction and improvements of ML and DL to the NIDS. The future trends for improving an anomaly-IDS for continuing detection in the evolution of cyberattacks are highlighted in several research studies.

翻译：基于人工智能（AI）的入侵检测系统（IDS）被视为在复杂网络中主动检测新型攻击的潜在机制。尽管现有综述论文采用系统综述或简单方法对异常网络入侵检测系统（NIDS）相关工作进行评估与批判，但当前综述仍沿用传统定量描述方式，通过综合与总结数据对比来发现现有空白，未考虑算法性能。本文针对网络入侵检测系统（NIDS）中的人工智能技术展开系统性元分析研究，重点聚焦网络安全领域的深度学习（DL）与机器学习（ML）方法。在结构层面解析深度学习算法，并基于网络基础设施与攻击类型论证数据入侵网络的合理性。通过进行元分析并从有效性、使用数据集、检测攻击类型、分类任务及时间复杂度等维度对DL与ML方法的验证效果进行探讨，我们为当前基于NIDS的出版物提供了一套基于系统方法的全面基准评估。该方法被视为对基于异常的入侵检测系统（anomaly-NIDS）模型的综述性研究。此外，针对ML与DL在NIDS领域的最新方向与改进，讨论了所提算法与所选数据集的有效性。多项研究强调了在网络攻击演进过程中持续检测的异常入侵检测系统（anomaly-IDS）未来改进趋势。