Computer Networks and Communications

DDoS Detection Using Machine Learning for Cloud Service Providers

2026-03-17T10:48:13+08:00

Distributed Denial of Service (DDoS) attacks pose severe threats to Cloud Service Providers (CSPs) due to their massive network scale and unique traffic characteristics. This paper proposes a comprehensive detection framework that addresses CSP-specific challenges through integrated Machine Learning (ML) models and visualization techniques. Our approach combines feature selection algorithms (Salp Swarm Algorithm, Gray Wolf Optimization, Particle Swarm Optimization) with ten Machine Learning and Deep Learning classifiers (Logistic Regression, K-Nearest Neighbors, Random Forest, AdaBoost, Support Vector Machines, Decision Trees, XGBoost, Naïve Bayes, Artificial Neural Networks, Long Short-Term Memory) optimized for CSP-scale traffic. Experimental validation is conducted using a hybrid dataset that combines the benchmark The Canadian Institute for Cybersecurity-IoT (CICIoT)-2023 dataset with real-world CSP backbone traffic, where 65% of the data is from real CSP environments. The proposed framework achieves high detection rates, with 99.9% accuracy and an AUC of 0.999. While these metrics are exceptional, we acknowledge that they represent performance on our specific hybrid dataset and may vary in real-world environments, particularly in the presence of zero-day attacks. The framework demonstrates high accuracy while addressing the "weak signal" problem inherent to hyperscale environments. Visualization components provide critical insights into feature correlations, attack distributions, and model performance trade-offs. This research extends traditional DDoS detection methods by incorporating bio-inspired optimization and comprehensive visualization, providing CSPs with actionable intelligence for real-time threat mitigation.

A Novel Soft Computing-Driven Entropy-Fuzzy Rule Optimization Framework for Early Adaptive Intrusion Detection

2026-03-10T11:46:19+08:00

This paper proposes the Entropy-Fuzzy Rule Optimized Synergistic Tuning Framework for Intrusion Detection System (E-FROST IDS), an entropy-driven, fuzzy-rule-optimization-enhanced intrusion detection system designed to address challenges related to model accuracy, flexibility, and uncertainty. The results suggest that E-FROST improves the effectiveness of anomaly detection, reduces false alarms, and enhances interpretability through entropy-based uncertainty modeling. The novelty lies in integrating fuzzy entropy with an efficient decision engine tailored for dynamic threat environments. The main contribution is a unified and adaptive IDS capable of identifying threats across a wide range of network conditions. Major implications include strengthened cyber-defense preparedness, greater transparency for security analysts, and a flexible framework for future intelligent intrusion detection system solutions.

Exploring Approaches for Secure Medical Image Storage and Retrieval: A Comprehensive Survey

2026-03-11T10:06:50+08:00

Medical images, the backbone of digital healthcare, hold critical information for accurate diagnoses and lead to advancements in medical research. However, the increasing amount of these images often entails storage on third-party cloud servers where security and privacy are major concerns. In digital healthcare operations, securely storing and retrieving medical images involves considerable difficulties. This paper investigates how existing techniques handle medical images while maintaining performance and security. It analyzes secure storage strategies like blockchain and encryption and covers secure retrieval strategies like indexing that preserve data security and effective search capabilities. The paper provides insights into the advantages, disadvantages, and applicability of current systems through an extensive evaluation. It also offers thoughtful suggestions and future research paths that could establish a trustworthy and secure medical image management.

Coded Caching in Combinatorial Multi-Access Networks with Private Caches

2026-03-09T10:01:23+08:00

Content caching at the network edge plays a critical role in reducing server load and improving scalability in large-scale content delivery networks. We consider a cache-aided multi-access network in which each user is equipped with a private cache and connects to a distinct subset of access caches. A central server stores the content library and populates both private and access caches using uncoded placement. For this setting, we establish upper and lower bounds on the optimal worst-case rate under uncoded placement, leveraging the Maddah-Ali-Niesen (MAN) schemes for dedicated and combinatorial multi-access caching, and also derive a cut-set lower bound under general placement. We then propose an extension scheme that integrates the principles of the MAN schemes for dedicated and multi-access caching, generalizing them to the considered setting, at the cost of high subpacketization complexity. To address this limitation, we introduce an improved coded caching scheme for a fixed private cache size, which achieves significantly lower subpacketization while maintaining comparable rates. Under the placement strategy of the improved scheme, we further derive an index-coding-based lower bound on the rate. Numerical comparisons demonstrate that the improved scheme reduces rate and subpacketization, with performance approaching theoretical lower bounds as the multi-access connectivity increases. In addition, we present a general placement policy applicable to arbitrary private cache sizes. Finally, we show that the proposed scheme is order-optimal in certain regimes and prove its optimality when the number of access caches equals four.

Technological Bibliography Study on UAV and IoT Wireless Communication Specification Through 5G Cellular Networks

2026-02-26T11:04:18+08:00

The growth of Information and Communication Technologies (ICTs) involvement in our daily life elicits currently a reliable network connection at anytime and anywhere. Robust cellular network is required for the approval of demand users' demand. In the late of last century, noticeable studies were invested in the ICT infrastructure development to meet the public and industrial needs. The evolution of mobile telecommunication has transitioned through multiple phases, from early voice-centric First-Generation (1G) systems to the data-heavy Fourth-Generation (4G) era. Currently, Fifth-Generation (5G) technology has emerged as a superior standard, prioritizing enhanced security protocols and high-speed data transmission according to nomad users. The global rollout of 5G frameworks is inherently linked to the integration of Internet-of-Things (IoT) ecosystems. Then, the design of relevant Base Stations (BSs) for communication during the urban social events constitutes a challenging problem of IoT infrastructures. Scholars have identified Unmanned Aerial Vehicles (UAVs) as a versatile solution for supplementing BS infrastructure, particularly for broadening the reach and throughput of 5G deployments. In scenarios involving public safety, UAV-integrated networks function as airborne base stations, facilitating advanced technologies such as millimeter-wave signaling, IoT connectivity, and 3D Multiple-Input-Multiple-Output (MIMO) arrays. The review on UAV-based wireless communication system modelling by taking into account the different classes and specification with respect to the regulation of technology is synthesized. The synthesis review of UAV cellular network planning for the urban cities is reported by stating about the experiment and performance evaluation. The research works highlight the technical challenge on the UAV-BS deployed for urban social events. It can be emphasized that the study of the UAV wireless communication necessitates the modeling consideration of channel from air transmission to ground BS reception, optimal deployment and the optimization of trajectory. To guarantee the communication quality, the analysis of system performance using UAVs is based on the energy efficiency and resource management.

Mirai Botnet Multi-Class Attack Detection Through Machine Learning and Feature Selection

2025-11-24T16:20:38+08:00

Machine Learning (ML), which provides timely insights for efficient threat identification and prevention, has become a crucial cybersecurity technology. However, the growing number of features in modern datasets increases both processing complexity and computational cost. By concentrating on feature selection and extraction techniques, this study seeks to improve the efficacy of Mirai botnet analysis. A data extraction approach that transforms Internet of Things (IoT) network attack datasets (in Packet Capture (PCAP) format) to flow-driven attributes (in Comma-Separated Values (CSV) format) was presented in our earlier work. A unique framework for effectively developing, assessing, and analyzing Mirai botnet assaults in IoT networks is provided by the obtained and labeled features of the Mirai-based multi-class IoT botnet dataset. In this study, experiments were conducted using the extended Mirai-based multi-class dataset and the widely used Network Security Laboratory-Knowledge Discovery in Databases (NSL-KDD) dataset for comparison. The results of both experiments demonstrate that Random Forest Feature Importance (RFFI) outperforms the Boruta feature selection algorithm. Furthermore, the random forest and decision tree models achieved superior performance in all tests, attaining 100% accuracy in the first experiment using the extended dataset. These findings highlight the importance of selecting relevant features, rather than using all available attributes, to enhance detection performance and computational efficiency.