Scientific Affairs and Research - IJSU

Just Published

Research Article in our journal (Mesopotamian journal of Cybersecurity) indexed in Scopus, CiteScore(7.7).

Title: Enhancing Advanced Persistent Threat Detection with Federated Learning and Neural Networks for Secure Cloud Computer Environment.
DOI: https://doi.org/10.58496//MJCS/2025/068
Keywords:
Advanced Persistent Threats (APTs), federated learning (FL), Exploratory Data Analysis (EDA), Gated Recurrent Unit (GRU) model, Convolution neural networks (CNN) model .
Highlights:
This paper proposes FedNN-APT, a federated learning–based intrusion detection system for identifying Advanced Persistent Threats (APTs) in distributed and resource-constrained cloud environments. The framework combines GRU and CNN models to capture temporal and spatial attack patterns while enabling privacy-preserving, decentralized training. Evaluated on an APT malware dataset, the hybrid GRU-CNN model achieves superior performance, reaching near-perfect accuracy across multiple clients. Results demonstrate that FedNN-APT outperforms existing APT detection methods and highlights the effectiveness of integrating deep learning with federated learning for secure and scalable cloud-based threat detection.

Just Published

Research Article in our journal (Mesopotamian journal of Cybersecurity) indexed in Scopus, CiteScore(7.7).

Title: A Novel of Bounded Zeghdoudi Distribution: Estimations, Simulation and Applications.
DOI: https://doi.org/10.58496//MJCS/2025/067
Keywords:
Zeghoudi distribution, Unit distributions, Moments , Inequality measures, Statistical inference .
Highlights:
This article introduces the Bounded Zeghoudi Distribution (BZD), a unit-interval extension of the Zeghoudi distribution that improves flexibility while preserving simplicity. The BZD accommodates various density shapes and hazard rate behaviors and its key statistical properties are derived. Parameter estimation is examined using sixteen methods with simulation validation. Applications to real proportional datasets, including cybersecurity-related indicators, show that the BZD provides better fit and performance than several well-known competing unit distributions.

Just Published

Research Article in our journal (Mesopotamian journal of Cybersecurity) indexed in Scopus, CiteScore(7.7).

Title: A Multi-Factor Quantum-Resistant and Privacy-Preserving Authentication Protocol for Decentralized Systems.
DOI: https://doi.org/10.58496//MJCS/2025/066
Keywords:
Quantum-resistant authentication, Privacy-preserving protocols, Lattice-based cryptography, Multi-factor authentication, Decentralized authentication .
Highlights:
This paper presents a quantum-resistant, privacy-preserving authentication protocol (PPAP) for decentralized systems. It combines lattice-based cryptography (ML-KEM) with privacy-enhancing techniques and CKKS-based multi-factor authentication to securely support biometric and signature-based factors. Formal analysis using cryptographic proofs and the Tamarin Prover confirms resistance to classical and quantum attacks. Performance results show low computational and communication overhead, with the Time-Aware Predictive Access Model (TAPM) further improving authentication efficiency.

Just Published

Research Article in our journal (Mesopotamian journal of Cybersecurity) indexed in Scopus, CiteScore(7.7).

Title: Arcsine Ratio Sine Generalized Distributions with Applications to Biomedical and Engineering Data.
DOI: https://doi.org/10.58496//MJCS/2025/065
Keywords:
Arcsine ratio sine generalized family, Identifiability , Critical points, Moments , Score function, Simulation .
Highlights:
This study explains a new class of trigonometric probability distributions called the Arcsine Ratio Sine Generalized (ARS-G) family. It presents explicit mathematical formulas for the statistical properties of this family. As an example, the study focuses on the ARS-Weibull (ARS-W) distribution, which is derived from the Weibull model. The ARS-W distribution is shown to be highly flexible, with hazard rate functions that can take several shapes such as bathtub, reversed bathtub, J-shaped, L-shaped, and unimodal forms. The study also explains how the model parameters can be estimated using different estimation techniques and evaluates their performance through Monte Carlo simulations. Finally, the study demonstrates the practical usefulness of the ARS-W distribution by applying it to real lifetime data, where it provides a good fit and, in some cases, performs better than classical distributions.

Just Published

Research Article in our journal (Mesopotamian journal of Cybersecurity) indexed in Scopus, CiteScore(7.7).

Title: A Proposed Algorithm For Avoiding Jammer In Structure-Free Wireless Sensor Networks.
DOI: https://doi.org/10.58496//MJCS/2025/064
Keywords:
Denial-Of-Service (Dos) Attacks., structure-Free .Jamming, Wireless Sensor Networks (Wsns), Signal-To-Jamming Ratio ) SJR) And Attack.
Highlights:
Wireless Sensor Networks (WSNs) are widely used in critical environments but suffer from limited, non-rechargeable energy resources, making them vulnerable to failures and jamming attacks. This study proposes a structure-free, adaptive transmission protocol that dynamically selects routing paths to improve energy efficiency and communication reliability. The protocol is evaluated against four jammer types—Constant, Deceptive, Random, and Reactive—using metrics such as energy consumption, signal-to-jamming ratio, and data rate. Simulation results show that jamming can increase transmission levels by up to 55% and energy consumption by up to 62%, with Random Jammers having the greatest energy impact. Data rates also rise by about 37% under jamming conditions. Overall, the results demonstrate that the proposed protocol effectively mitigates jamming effects while conserving energy, making it suitable for WSNs operating in hostile or high-risk environments.

Just Published

Research Article in our journal (Mesopotamian journal of Big Data) indexed in Scopus, CiteScore(6.9).

Title: Shared Generator-Based Serverless Multimodal Federated Learning For Medical Image Analysis.

DOI: https://doi.org/10.58496/MJBD/2025/015
Keywords:
FL, SGS-FL, Medical imaging, Lung nodules, Multimodality models.
Highlights:
This paper presents SGS-FL (Shared Generator–Serverless Federated Learning), a decentralized multimodal framework for medical image analysis, addressing privacy, data scarcity, and heterogeneity challenges in centralized AI model development. SGS-FL uses a shared generator with multiple discriminators to eliminate central server dependency and employs latent space aggregation with attention and independent component analysis to enhance interpretability, fairness, and feature relevance. Evaluated on three lung cancer datasets—LIDC-IDRI (CT scans), NODE21 (X-rays), and NSCLC radiogenomic PET-CT—the model achieved 92.5% ± 1.2 accuracy, 0.83 ± 0.02 Dice coefficient, 0.946 ± 0.01 AUC, and a 21.5 ± 1.1 FID, outperforming state-of-the-art federated methods such as FedACS and Federated Transfer Learning by a statistically significant margin (p < 0.01). SGS-FL demonstrates high scalability, interpretability, and robustness, establishing a privacy-preserving and clinically reliable AI paradigm for medical imaging.

Just Published

Research Article in our journal (Mesopotamian journal of Big Data) indexed in Scopus, CiteScore(6.9).

Title: A Hybrid Machine Learning Approach for Enhanced Diabetes Prediction: Integrating Image and Numerical Data.

DOI: https://doi.org/10.58496/MJBD/2025/014
Keywords:
Diabetes Mellitus (DM), Machine Learning (ML), Deep Learning (DL), Multimodal Data Integration.
Highlights:
This study proposes a hybrid machine learning model for early and accurate prediction of diabetes mellitus (DM) by combining deep learning on retinal images with gradient-boosting machines (GBMs) on clinical and demographic data. The model integrates data from the Pima Indians Diabetes Database and the APTOS 2019 fundus image dataset, applying standardized preprocessing and feature ranking. By fusing visual and tabular features into a shared latent representation, the approach captures both systemic and localized disease indicators. Experimental results show superior performance over single-modality models, achieving 96% accuracy, 0.96 macro F1-score, and 0.994 AUC-ROC. This multimodal framework enhances diagnostic robustness and supports more reliable clinical decision-making, demonstrating the potential of integrated AI approaches for complex disease prediction.

Just Published

Research Article in our journal (Mesopotamian journal of Big Data) indexed in Scopus, CiteScore(6.9).

Title: Towards Autonomous Optical Fibre Networks: High-Precision EDFA Gain and Spectral Response Prediction via Hybrid CNN-LSTM Deep Learning.

DOI: https://doi.org/10.58496/MJBD/2025/013
Keywords:
Hybrid CNN-LSTM, EDFA gain prediction, Real-time optimization, Spatiotemporal modelling, Optical amplifier control
Highlights:
The study proposes a hybrid CNN–LSTM model for automatic prediction and optimization of erbium-doped fiber amplifiers (EDFAs) in long-distance optical networks. The model combines spatial–spectral analysis from CNNs with temporal dynamics from LSTMs to predict gain characteristics and bandwidth. It achieves very high accuracy (R² > 0.999) and an ex*****on time of only 6.1 milliseconds, making it 600,000× faster than traditional methods and up to 88% more efficient, enabling real-time and autonomous EDFA optimization.

Just Published

Research Article in our journal (Mesopotamian journal of Big Data) indexed in Scopus, CiteScore(6.9).

Title: AI-Driven Smart Contract Vulnerability Detection: A Systematic Review of Methods, Challenges, and Future Prospects.

DOI: https://doi.org/10.58496/MJBD/2025/012
Keywords:
Artificial Intelligence, Sustainable Growth, Security analysis, Deep Learning, Machine Learning, Smart Contracts, Blockchain, Vulnerability detection, Ethereum, Sustainable Digital Infrastructure, Sustainable AI-Blockchain Integration
Highlights:
This study provides a systematic review of smart contract (SC) vulnerability detection methods published between 2020 and 2024, analyzing 21 key studies to evaluate current progress and identify gaps in smart contract security research. Smart contracts, while enabling automated, trustless transactions on blockchains, remain highly vulnerable to security flaws that can lead to significant financial loss. The review categorizes detection approaches into static analysis, dynamic testing, AI-driven models, graph-based techniques, and hybrid systems, comparing their effectiveness, scalability, and practical deployment challenges. Findings show that AI-based methods, particularly those using deep neural networks (DNNs) and graph neural networks (GNNs) (e.g., ContractWard with 98.48% Micro-F1 and SCVDIE-ENSEMBLE with 95.46%), offer high accuracy but require substantial computational resources, limiting use in constrained environments. Conversely, lightweight tools such as Slither and NeuCheck provide faster, more efficient analysis but detect fewer complex vulnerabilities. Emerging real-time monitoring systems like SODA and GPTScan show promise in reducing false positives and enabling proactive threat detection. However, major challenges persist, including reliance on labelled datasets, limited generalization to novel attacks, and scalability constraints. The review underscores the need for hybrid, adaptive frameworks that combine efficiency, accuracy, and robustness for practical, real-world smart contract security.

Just Published

Research Article in our journal (Mesopotamian journal of Big Data) indexed in Scopus, CiteScore(6.9).

Title: Hybrid Quantum-Assisted Deep Learning Model for Early-Stage Alzheimer’s Disease Classification Based on MRI Images.

DOI: https://doi.org/10.58496/MJBD/2025/011
Keywords:
Alzheimer's disease, Quantum machine learning, Medical image classification, Quantum Fourier transform, Hybrid neural networks
Highlights:
This study presents HQC-Net, a hybrid quantum–classical neural network for precise four-class classification of Alzheimer’s disease (AD) using MRI scans. By integrating classical deep learning (custom CNN and modified ResNet18) with six-qubit variational quantum circuits employing multiaxis rotation encoding, a quantum Fourier transform, and multihead attention, HQC-Net leverages quantum computing’s ability to process complex high-dimensional features. Evaluated on the Kaggle (5,121) and OASIS (20,000) datasets under realistic quantum noise conditions, the model achieved 99.67% test accuracy and AUC = 1.0000, outperforming other quantum-enhanced approaches by 3.57% and demonstrating exceptional sensitivity in detecting very mild dementia (99.86% accuracy). The results highlight HQC-Net’s practical quantum advantage for early Alzheimer’s diagnosis, combining superior accuracy, robustness, and efficiency suitable for current Noisy Intermediate-Scale Quantum (NISQ) hardware.