The article describes a model of cloud-native AI pipelines designed for continuous optimization of computing infrastructure and real-time anomaly detection. The developed model combines modern approaches to observability, machine learning (ML), and auto-scaling   based   on   load forecasting.  The methodology is based on the use of LSTM models, autoencoders, and convolutional neural networks (CNN) integrated into Kubernetes environment with support for Prometheus, Kafka, and Grafana. Load changes are simulated, and the system's response to critical events is evaluated. The results demonstrate a significant improvement in anomaly detection accuracy (up to 93%) and resource efficiency (up to 26% cost reduction compared to traditional approaches). The proposed model can be used in AIOps systems that require a high level of automation and reliability.

Mobile Edge Computing (MEC) mitigates cloud computing systems’ latency and limited responsiveness by offloading computationally intensive tasks from user devices to nearby Edge Servers (ESs). However, achieving efficient offloading under dynamic mobility, fluctuating link quality, and constrained resources remains a significant challenge. To address this, we propose MSQ, a lightweight and adaptive three-dimensional decision offloading model that jointly incorporates Mobility, Sociality, and QoS awareness. MSQ employs Kalman filtering for mobility prediction, Rényi entropy to quantify social affinity among mobile users, and Affinity Propagation (AP) clustering to reduce redundant ES candidates while balancing computational load. Comprehensive experiments across small and medium-scale MEC networks demonstrate that MSQ reduces average task delay by up to 78%, energy consumption by 66%, and load imbalance by 64% compared with a random offloading strategy while having decision latency below one millisecond. Moreover, MSQ lowers the 95th-99th percentile tail delays by 35-45%, ensuring smoother and more reliable user experience in real- time applications. These results confirm that MSQ offers a scalable, low-latency, and energy-efficient offloading decision suitable for dynamic and intelligent edge systems.

To address the growing multi-user interference in dense wireless networks, we propose an interference-aware Deep Quantum Neural Network (DQNN) for channel estimation in the Non-orthogonal multiple access (NOMA) systems. The proposed method incorporates a hybrid classical-quantum architecture. A Transformer-encoder processes the pilot signals to extract spatiotemporal features. A parameterized quantum circuit maps the processed features into a high-dimensional Hilbert space. The enhancement hinges on an Adaptive Energy Valley Optimization (AEVO) algorithm, which modifies the optimization trajectory using interference-aware preconditioners derived from the interference covariance structure. With the aid of these preconditioners, the DQNN can steer through the NOMA's non-convex terrain characterized by interference to enhance estimation performance.  Moreover, interference-aware preconditioning is achieved through a lightweight neural network which adapts to time-varying interference. The successive interference cancellation decoder uses the estimated channel matrix to recover symbols. By further analysing the results, it is noticed that the quantum-enhanced machine learning delivers better results than the classical ones. The proposed framework enhances the state-of-the-art in NOMA channel estimation, while also providing a general framework for interference-aware optimization in quantum machine learning. At 10 dB SNR, the AEVO-DQNN method with a 16x16 antenna array obtained a minimum NMSE of 0.012288 and a minimum BER of 0.013023. Further, the proposed method outperforms the competing methods in terms of NMSE/BER mean with 95% confidence intervals, interference rejection ratio analysis, sensitivity to estimation error and estimated interference covariance, and paired t-test analysis.

With the rapid increase in malware threats, robust classification methods have become essential to protect digital environments. This study conducts a comparative analysis of machine learning and deep learning methods for malware detection. A variety of models are used from both machine learning and deep learning paradigms to determine their effectiveness in distinguishing malware. To further refine the models, several feature selection techniques are applied to reduce the dimensionality of the data and enhance performance. Performance metrics, including accuracy, precision, recall, and F1-score is used to evaluate each model. The findings indicate that while deep learning approaches generally provide higher detection accuracy, feature selection methods contribute significantly to improving machine learning models in terms of performance and computational efficiency. This analysis offers valuable insights into the balance between model complexity and effectiveness, providing practical recommendations for implementing malware classification systems in real-world applications.

The increase in cyber attacks leads to significant challenges to the security in Fog based IoT environments. Existing studies have been implementing machine learning (ML), ensemble learning (EL) and deep learning (DL) for security, in this study we opted deep ensemble learning (DEL) for detection of threats in fog based IoT environments. The proposed DEL model is build using Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Gated Re-current Units (GRU) as base models, and it is augmented using a metalearner such as Logistic Regression (LR), Random Forest (RF), AdaBoost, XGBoost, CatBoost, and LightGBM and also with a Voting Classifier (VC) for f inding the best model. In our experimentation, the evaluation is performed with different datasets such as DDoS SDN, NSL-KDD, UNSW-NB-15, and IoTID20. In this work, DEL with RF achieved better performance than other models when performance metrics such as accuracy (Acy), precision (Prn), recall (Rcl), F1-Score (F1-S) and AUC-score are considered. For instance, DEL with RF achieved an Acy of 99.99%, Prn of 100%, Rcl of 99.96%, F1-S of 99.98% and AUC-score of 1.00 on IoTID20 dataset. Afterward, to analyze the network performance of the DEL models at fog, we have considered the metrics such as cost, energy, resource utilization, latency and service time. This work shows that DELmodels can improve the security of fog assisted IoT systems.

With the rapid development of mobile technologies, analyzing data generated by mobile devices is becoming increasingly important. A wide range of applications, from marketing to healthcare, require the development of effective methods for extracting valuable information from this data. This study is devoted to developing a methodology for assessing an individual's social capital based on the analysis of mobile communication data. To assess social capital, we propose a two-stage Cascade Model that combines the advantages of the Random Forest (RF) and Logistic Regression (LR) algorithms. In the first stage, RF is used to select the most significant features reflecting various aspects of social capital. In the second stage, LR is used to assess of social capital, taking into account nonlinear relationships between features. The results of the study open up new opportunities for studying social phenomena and can be used in as sociology, marketing, and urban planning.

Wireless Sensor Networks consist of energy constrained sensor nodes that monitor and transmit data to a central base station. These networks are highly susceptible to link and node failures, which further degrades performance and reduce overall network reliability. In this paper we have addresses these challenges and proposed a reinforcement learning based self-healing routing (RL-SHR) protocol, implemented in NS2 simulation environment. In the work, each node functions as an autonomous RL agent that learns optimal routing paths by interacting with the network environment and adapting to failure conditions. The protocol enables nodes to dynamically avoid unreliable paths, recover from faults, and optimize performance over time. Simulation results shows that the proposed protocol significantly outperforms traditional routing protocols such as AODV and DSR in terms of packet delivery ratio, end-to-end delay, energy consumption and network lifetime under varying failure scenarios. This work lays the groundwork for integrating learning based resilience mechanisms into next generation sensor networks.

Due to the dynamic nature of the network architecture, resource constraints, and susceptibility to security attacks, securing data transmission in Mobile Ad-hoc Networks (MANETs) is a significant problem. This work proposes a novel Equivariant Quantum Neural Networks with Adaptive Tangent Brakerski-Gentry Vaikuntanathan Homomorphic Encryption algorithm (EQNN-ATBGVHEA)- based secure routing in MANET. The suggested approach comprises three steps: cluster head (CH) selection, optimal path selection, and secure data transfer. Initially, the Bowerbird Optimization Algorithm chooses the CH and sends the message through the constructed path. Once the clusters are established, data is transferred between the sender and receiver. For optimal route selection, developed the EQNNs technique which incorporates a neural network for quick route selection. EQNN resolves the issues of local optimality by constructing a new fitness process based on residual energy (RE) and delay. After the optimal path selection, Data transfer is secured by the innovative ATBGVHEA technique. Furthermore, this method is built using NS3, and the variables are determined. Additionally, the acquired results are contrasted with existing approaches for validating the efficiency of the suggested strategy. The developed method achieved a clustering accuracy of 98.5%, a computational time of 55ms, and a residual energy of 0.44.

An innovative approach to identifying rapidly spreading false information is to create a targeted graph and its subsequent clustering. A method for detecting rapidly spreading fake messages in social networks has been developed. K-means, Louvain, and Leiden algorithms were applied to identify large communities in graphs, enabling the rapid detection of fake news. A modified fake news detection algorithm based on k-means and Leiden can group fake news clusters, enabling rapid identification of widely spreading news. The combination of Leiden for structural analysis of communities and SVM for classification provides an optimal balance between accuracy (F1-score = 0.87) and completeness of fake detection (Recall = 97%), allowing the system to be used both for analysing large datasets and for monitoring new publications. The Lei-den algorithm demonstrated the highest modularity (Q = 0.7212), which is 4.8% better than Louvain (Q = 0.6884), and detected 40 structural communities. The modified method has a lower modularity (Q = 0.5584), since modularity is not calculated for K-means.

Device-to-Device (D2D) communications in 5G enabled vanet Networks offer significant advantages in terms of improved communication efficiency and reduced latency. However, ensuring secure and efficient key agreement among devices remains a critical challenge. In this study, we present a novel lightweight framework for D2D communications that addresses these concerns by employing a Signcryption-based key agreement scheme [1]. The proposed scheme is built on the foundation of Diffie-Hellman Hyper Elliptic Curve Cryptography and leverages two one-way cryptographic hash functions to enhance security. By integrating the signcryption technique, our framework achieves a seamless combination of encryption and signing [2], reducing computational overhead and conserving network resources in resource-constrained 5G-enabled devices. Furthermore, we prioritize user location privacy in our framework by employing advanced techniques, including the Chinese Remainder Theorem. This ensures that location information is protected and not exposed to unauthorized parties during D2D communication sessions. Through extensive simulations and performance evaluations using ns3, we demonstrate the effectiveness and efficiency of our proposed key agreement scheme for D2D communications in 5G enabled vanet Networks. The results show improved communication performance and reduced resource consumption, making our framework a promising solution for secure and efficient D2D interactions in the context of evolving 5G networks.

The rapid extension of the Internet of Things (IoT) has introduced significant concerns, particularly in ensuring data security and safeguarding sensitive and private data. The integration of Federated Learning into IoT architecture has occurred as a covenanting solution to address the risks of data breaches, resource efficiency, and the challenges of data privacy and security. This paper presents a novel lightweight framework tailored for resource-constrained IoT devices that integrates Federated Learning and Tiny Machine Learning (TinyML) to deploy lightweight, reliable models on edge devices. Our experimental results show that the proposed approach can improve efficiency, reduce communication overhead, and enhance privacy preservation.

A Wireless Sensor Network (WSN) is an efficient system for monitoring distributed areas and controlling environments; however, such networks are susceptible to malicious node attacks that bring forth network insecurity and untrustworthy data. WSNs are vulnerable to malicious nodes and cyber attackers that can interfere with data transmission, leading to compromised decision-making systems. Traditional security techniques against WSNs lack flexibility in real-time detection and data integrity because of constrained processing resources and vulnerabilities from centralized storage. This work aims to improve detection accuracy through a multi-stage strategy, which constitutes the general objective of this research. The presented model uses WSN-DS and WSN-BFSF datasets. The data are pre-processed using Localized-Global Depth Normalization for uniformity, followed by feature selection via Boosted Tern-Cat Hunting Optimization, which combines Cat Hunting Optimization and Boosted Sooty Tern techniques to reduce dimensionality. The attack detection is performed by a Parallel Triple Graph Attention-based Convolution Network, which employs Quantum Parallel Deep Convolution and Triple Graph Attention Networks. The RMRO optimizes the model's parameters to classify more accurately, and the benign data are safely stored through the Consensus-Aided PoA Decision Blockchain Engine and InterPlanetary File System. This approach achieved 99.4% accuracy, 99.3% recall, and 99.5% F1 score on the WSN-DS dataset and 99.2% accuracy, 99.1% precision, and 99.3% F1 score on the WSN-BFSF dataset while showing robustness across different combinations of sensors. Hence, the Tri-QPdCNet offers a pioneering approach toward securing WSNs from dynamic and persistent attacks by providing an improved framework for anomaly detection using a strong, scalable architecture, augmented with blockchain technology. That leads to more robust WSN infrastructures that can be more securely and smoothly deployed in real-time critical environments.