A kind of wireless network called a “mobile ad hoc network” (MANET) can transfer data without the aid of any infrastructure. Due to its short battery life, limited bandwidth, reliance on intermediaries or other nodes, distributed architecture, and self-organisation, the MANET node is vulnerable to many security-related attacks. The Internet of Things (IoT), a more modern networking pattern that can be seen as a superset of the paradigms discussed above, has recently come into existence. It is extremely difficult to secure these networks due to their scattered design and the few resources they have. A key function of intrusion detection systems (IDS) is the identification of hostile actions that impair network performance. It is extremely important that an IDS be able to adapt to such difficulties. As a result, the research creates a deep learning-based feature extraction to increase the machine learning technique's classification accuracy. The suggested model uses outstanding network-constructed feature extraction (RNBFE), which pulls structures from a deep residual network's many convolutional layers. Additionally, RNBFE's numerous parameters cause a lot of configuration issues because they require manual parameter adjustment. Therefore, the integration of the Rider Optimization Algorithm (ROA) and the Spotted Hyena Optimizer (SHO) to frame the new algorithm, Spotted Hyena-based Rider Optimization (S-ROA), is used to adjust the RNBFE’s settings. Attack classification is performed on the resulting feature vectors using fuzzy neural classifiers (FNC). The experimental analysis uses two datasets that are publicly accessible.

一种称为“移动自组织网络”（MANET）的无线网络可以在不借助任何基础设施的情况下传输数据。由于电池寿命短、带宽有限、依赖中介或其他节点、分布式架构和自组织，MANET 节点容易受到许多与安全相关的攻击。物联网（IoT）是一种更现代的网络模式，可以被视为上述范式的超集，最近已经出现。由于这些网络的设计分散且拥有的资源很少，因此保护这些网络的安全极其困难。入侵检测系统 (IDS) 的一个关键功能是识别损害网络性能的敌对行为。 IDS 能​​够适应这些困难是极其重要的。因此，该研究创建了基于深度学习的特征提取，以提高机器学习技术的分类准确性。建议的模型使用出色的网络构造特征提取（RNBFE），它从深度残差网络的许多卷积层中提取结构。此外，RNBFE 的众多参数导致了很多配置问题，因为它们需要手动调整参数。因此，结合骑手优化算法（ROA）和斑点鬣狗优化器（SHO）来构建新算法——基于斑点鬣狗的骑手优化（S-ROA），用于调整RNBFE的设置。使用模糊神经分类器 (FNC) 对生成的特征向量进行攻击分类。实验分析使用两个可公开访问的数据集。