The rapid increase in the use of Internet of Things (IoT) devices has transformed everyday life and industries such as healthcare, transportation, and smart homes. However, these devices, often limited in resources, depend on communication across edge, fog, and cloud layers, creating vulnerabilities that attackers can exploit. This paper provides a comprehensive review of intelligent intrusion detection system (IDS) tailored for IoT security, focusing on solutions that utilize Artificial Intelligence (AI), Machine Learning (ML), and Deep Learning (DL). We analyze existing IDS approaches for IoT devices and secure communication across the edge, fog, and cloud layers, highlighting their strengths and limitations. Additionally, we identify Key research challenges include computational complexity, real-time adaptability, and energy efficiency in Edge Computing. To address these gaps, we propose future research directions, including neuromorphic computing for ultra-fast IDS, self-evolving AI-driven IDS, hyper-personalized anomaly detection, federated learning for privacy- preserving security, and explainable AI (XAI) for human–AI collaboration. By integrating these innovations, we envision next-generation IDS solutions that offer scalable, interpretable, and energy- efficient security frameworks for the dynamic IoT ecosystem.

The rapid integration of Internet of Things (IoT) devices into various domains has led to heightened concerns regarding cybersecurity threats. IoT networks, characterized by their heterogeneous and resource-constrained nature, face numerous challenges in effectively detecting and mitigating intrusions. Consequently, there is a critical need for lightweight Intrusion Detection Systems (IDS) tailored for these environments. This paper introduces a novel lightweight IDS designed specifically for IoT environments. Our approach utilizes a mean thresholding-based Fast Correlation-Based Filter (mFCBF) algorithm for feature selection, enhancing efficiency while maintaining detection accuracy. Leveraging the power of LightGBM and XGBoost for classification, our proposed IDS demonstrates exceptional effectiveness in identifying malicious activities within IoT networks. Through extensive experimentation on benchmark datasets such as CICIDS 2017, NSL-KDD and BoT-IoT we validate the efficacy of our approach, achieving 99.34% accuracy for CICIDS 2017, 99.62% for NSL-KDD and 99.75% for BoT-IoT while maintaining an average CPU consumption of just 2.03% and Memory usages of approximately 1.2%. These results demonstrate that our IDS is both lightweight and robust, capable of effectively safeguarding against emerging threats.

Sound event detection (SED) plays a pivotal role in applications ranging from surveillance to healthcare, yet achieving high accuracy in dynamic acoustic environments remains challenging. This paper proposes an enhanced framework for SED by integrating the You Only Hear Once (YOHO) Convolutional Neural Network (CNN) architecture with a Bottleneck Attention Module (BAM) to improve feature localization and boundary detection in audio signals. Leveraging the Urban Sound Detection dataset, raw audio is preprocessed into mel spectrograms to capture spectral-temporal characteristics, which are fed into two models: (1) a baseline YOHO-based CNN and (2) a YOHO variant augmented with BAM. The attention mechanism dynamically weights channel-wise features, enabling the model to focus on salient audio regions while suppressing noise. Experimental results demonstrate that the attention-enhanced model achieves 90% test accuracy, outperforming the baseline YOHO model (87%) and exhibiting faster convergence. Key contributions include a systematic comparison of segmentation-by-classification approaches, a novel integration of BAM into YOHO for SED, and empirical validation of attention mechanisms in improving robustness against acoustic variability. This work underscores the potential of lightweight attention modules to enhance real-time sound event detection in resource-constrained loT and edge computing applications, while future directions explore temporal modeling and multimodal fusion for further gains.

Network Intrusion Detection Systems (NIDS) play a crucial role in safeguarding network infrastructures against evolving cyber threats. Traditional machine learning (ML) models often struggle with the multi-class classification of attacks due to limitations in capturing complex feature dependencies and handling imbalanced data distributions. In this study, we propose an ensemble multi-model approach to address these challenges and enhance the accuracy and robustness of intrusion detection systems. Leveraging deep learning (DL) techniques and ensemble learning methods, our approach aims to improve classification accuracy by effectively capturing intricate feature dependencies and mitigating data imbalance issues. We conduct experiments using NSL-KDD and CICIDS-2017 datasets, employing rigorous evaluation procedures such as cross-validation and comparative analysis with existing models.

This book chapter presents a comprehensive study on the development and implementation of a reliable resource optimization model for cloud computing using an adversarial neural network (ANN). Optimization and efficient resource allocation have become crucial with the increasing adoption of cloud computing to ensure optimal performance and meet user demands. This chapter addresses these challenges by proposing a unique approach that leverages the capabilities of ANN to optimize resource allocation in cloud environments. The chapter begins with an introduction to cloud computing and its significance in modern IT infrastructures. It emphasizes the need for effective resource allocation strategies to maximize resource utilization while adhering to service level agreements (SLAs). The limitations of existing resource allocation models are discussed, highlighting the necessity for a more reliable and efficient solution. The proposed model introduces a pioneering architecture founded on an adversarial neural network, which comprises a generator network and a discriminator network. The generator network is responsible for generating resource allocation plans, while the discriminator network assesses the quality of these plans based on predefined metrics. By means of an adversarial training process, the generator network acquires knowledge and expertise in generating optimized resource allocation strategies that surpass the capabilities of the discriminator network, thus resulting in improved reliability and performance. The chapter provides detailed insights into the system design and working of the ANN-based resource optimization model. It discusses the architectural considerations, hyperparameters, and training methodology employed. Furthermore, it addresses the challenges associated with training the ANN, such as mode collapse and training instability, and presents effective strategies to mitigate these issues. Various optimization algorithms and loss functions are explored to ensure efficient convergence and the generation of high-quality resource allocation plans. To evaluate the efficacy of the proposed model, extensive experimental evaluations will be conducted. Performance benchmarks will be established, and comparisons will be made against conventional resource allocation approaches. The experimental results will demonstrate the superior reliability and optimization achieved by the ANN-BPSO-RF model. Furthermore, the model will exhibit remarkable adaptability to changing workload demands and showcases its scalability in large-scale cloud environments.

This paper targets developing a Deep learning (DL) method for classifying encrypted traffic by utilizing FlowPics, which are visual representations of network traffic data. We leverage Convolutional Neural Networks (CNNs) to exploit the spatial features within FlowPics. However, traditional CNNs might not fully address channel-wise importance within feature maps. To address this, we incorporate Squeeze and Excitation (SE) blocks. SE blocks dynamically recalibrate feature map channels, emphasizing informative ones critical for encrypted traffic classification and suppressing less relevant ones. This targeted focus leads to improved performance compared to a simple CNN architecture. Our evaluation demonstrates robust performance across various encryption types (Non-VPN, VPN, and Tor) with minimal computational overhead. This work contributes to the field by showcasing the effectiveness of DL with CNNs enhanced by SE blocks for encrypted traffic classification.

The advancement of virtual networks such as cloud computing, Software-Defined Networking (SDN), and Network Functions Virtualization (NFV) has enabled seamless access to information globally, providing on-the-go services and improving operational efficiency. However, these advancements also present significant threats to the security of information and data on these virtual networks due to their interconnected nature and large attack surface. Among these threats, Distributed Denial-of-Service (DDoS) attacks are one of the most frequent threats to network security that cause severe damage, which hamper network performance. In a DDoS attack, the attacker overwhelms network resources on the victim's side by sending a massive number of packets or connection requests to the server using bots, thereby disrupting the server's functionality. The existing works explore only single supervised learning algorithms or a comparison of multiple individual classifiers for DDoS attack detection, which limits their effectiveness in handling diverse attack patterns. To address this limitation, our work proposes an ensemble learning approach that combines multiple supervised machine learning techniques to combat DDoS attacks. Specifically, our ensemble learning framework integrates Random Forest, Support Vector Machine (SVM), and AdaBoost techniques to detect and identify different DDoS attacks effectively.

In an era marked by the proliferation of interconnected devices, the Internet of Things (IoT) has emerged as a revolutionary technological paradigm. IoT networks encompass a vast array of devices, from smart appliances to industrial sensors, revolutionizing industries and everyday life. However, this ubiquitous connectivity has ushered in a new frontier of security challenges, necessitating the deployment of robust Intrusion Detection Systems (IDS). This paper presents a pioneering Multi-Method Stacked Feature Selection (M2SFS) approach-based IDS tailored explicitly for IoT networks. By orchestrating a diverse ensemble of feature selection techniques, the M2SFS framework adeptly curates the feature set derived from the CICIDS 2017 dataset. This comprehensive selection strategy optimally reduces dimensionality, mitigating the resource constraints inherent to IoT ecosystems. The proposed IDS, rooted in the M2SFS approach, demonstrates exceptional accuracy and efficiency in detecting anomalous network behaviors. By harnessing the collective strength of stacked feature selection, this IDS capitalizes on the synergistic potential of multiple feature selection methods. The result is an IDS uniquely adapted to the intricacies of IoT networks, effectively fortifying their security posture. Through rigorous experimentation using the CICIDS 2017 dataset, the proposed approach affirms the superiority of the M2SFS-based IDS over conventional methods. This study underscores the pressing need for IDS in IoT environments and highlights the promise of our innovative approach in safeguarding IoT networks against emerging threats.