AlKadhim Journal for Computer Science

Hybrid local intensity variation and edge features map-based multi-focus image fusion using Genetic algorithm

2026-03-03T12:11:02+03:00

This study proposes a multi-focus image fusion approach that considers Genetic Algorithm (GA) optimization to achieve the selection of the most appropriate fusion weights that are used to increase the quality of the resulting fused image. local intensity variations with a standard deviation filter used to extract texture features, edge detection with the Sobel operator is defined, and a variance feature is also determined. The optimized weights are used to identify the best combination of feature maps of these three extracted features to achieve an accurate fusion process. Experiments demonstrate that this approach successfully retains texture, edges, and variation, resulting in a fused image with improved visual quality and information richness.

Deep Learning Models In The Early Brediction Of Epileptic Seizures

2026-03-13T17:17:26+03:00

Abstract

Prediction of early epileptic seizures is still a challenge in clinical neuroscience due to the unpredictability of seizures and their psychological and physical dangers to the patient. EEG can be used as an important modality for seizure warning systems with the goal of improving patient quality of life. Hence, the paper will compare and evaluate the performance of several deep learning models for seizure prediction 15 minutes in advance using EEG data. To this end, a uniform experimental protocol was followed in order to ensure a comparison between models: LSTM, BiLSTM, Graph Neural Network, and Transformer. Models were evaluated based on the CHB-MIT database that consisted of long-term recordings from 24 patients. Each signal was subjected to different pre-processing steps such as filtering, normalization, and time window segmentation. Then, an individual set of features was extracted, while data between pre-ictal and ictal classes were made balanced. Models were studied based on various metrics such as accuracy, specificity, reproducibility, F1 coefficient, and area under the curve.

Therefore, the results from the deep learning models used show converging performance. The LSTM and BiLSTM models showed high accuracy in positive cases, with a retrieval rate of 0.99. The Transformer model showed the highest accuracy of 0.95 and highest specificity of 0.76. The Graph Neural Network (GNN) showed comparable performance to the other models, with competitive performance but not outstanding in any of them. The use of Explainable AI (XAI) enabled the assessment of the importance of the temporal phases and features in both the performance and the decisions of the model. Hence, the findings confirm that the Transformer model has a high capacity for early seizure prediction, and thus, it is useful in the development of reliable early warning systems in patients with epilepsy.

Developing an Enhanced IHBO Algorithm with Chaotic Binary Hashing to Optimize BiGRU for Industry 4.0 IIoT Anomaly Detection

2026-01-08T01:36:42+03:00

The rapid expansion of the Industrial Internet of Things (IIoT) in Industry 4.0 has increased the need for reliable anomaly detection to protect industrial services and critical operations. Deep learning–based IDS solutions can achieve strong performance, yet practical deployment may be affected by tuning effort and suboptimal search behavior when optimization is trapped in local optima, particularly when feature selection and model optimization are handled as separate stages. This paper develops an enhanced Improved Honey Badger Optimization (IHBO) algorithm and integrates it with a Bidirectional Gated Recurrent Unit (BiGRU) model for IIoT anomaly detection. To strengthen exploration and maintain population diversity during optimization, the proposed IHBO incorporates a logistic chaotic map as a controlled diversification mechanism. In addition, chaotic binary hashing is used to map continuous candidate representations into discrete binary decisions, enabling effective wrapper-based feature selection within the same optimization framework. The main novelty lies in performing integrated feature selection and BiGRU optimization within a single framework rather than treating these steps independently. Experiments on two benchmark datasets—an industrial IIoT dataset and UNSW-NB15—show that the proposed BiGRU–IHBO approach achieves 94.70% accuracy on the IIoT dataset and 99.29% accuracy on UNSW-NB15, with consistent improvements in precision, recall, and F1-score compared with baseline models reported in the same experimental setting.

Computer Science AI-Driven Performance Analytics for Educational Staff: An Empirical Study in Iraqi Schools

2026-03-03T12:47:00+03:00

This quantitative study investigates the impact of AI-based decision support systems on teacher performance and administrative efficiency in Iraqi secondary schools. A structured questionnaire was administered to 350 teachers and administrators across four governorates (Sulaimaniyah, Nineveh, Baghdad, and Al-Nasr District). Data were analyzed using SPSS v25 through reliability tests (Cronbach's alpha), validity tests (Pearson correlation), descriptive statistics, and regression analysis. The results revealed a strong positive correlation between AI-based systems and teacher performance (R = 0.939, R² = 0.875, p < 0.001), indicating that 87.5% of the variance in teacher performance is explained by AI system use. Additionally, AI tools significantly improved administrative planning efficiency (R² = 0.694) and decision quality (R² = 0.570). Staff acceptance of AI technologies also showed a substantial impact on institutional performance (R² = 0.642). The study recommends promoting AI integration in schools, enhancing technological infrastructure, and providing training programs for educational staff. These findings offer evidence-based insights for policymakers seeking to modernize educational management through smart analytics.

Intelligent Decision Degradation Analysis under Extreme Data Scalability in Enterprise Information Systems

2026-03-03T13:03:41+03:00

This paper provides a computer framework describing a research on the deterioration of the decision making of AI based enterprise information systems in reaction to the extreme data scale issues. The pattern of erosion of the decision-making accuracy of four AI architectures, that is, neural networks, random forests, support vector machines, and ensemble approaches, was studied by means of controlled multi-scenario simulation. These experiments were performed in four cases of scalability of linear growth, exponential burst, step-wise expansion and random volatility.

One of the most important innovations in our approach is a simulation engine which will be employed in order to measure degradation using various performance measures i.e. decision accuracy, response time, uniformity and computational load. We marked important threshold limits in our analysis, where the reliability of the system will be very low and where an early warning of the looming collapse would be detected. We also came up with predictive models that we used to predict degradation patterns.

The results show that the ensemble methods are much stronger with the average accuracy of 94.3 percent even in extreme stress conditions, which are not present in traditional architectures. On the other hand, exponential bursts exhibit the greatest performance disparities, and after their thresholds are exceeded, at least 25 performances are required to initiate execution. Finally, this work provides new techniques for assessing recovery dynamics, resilience, and decay rates. We provide organizations that are having trouble simplifying their architectures with practical, hands-on guidance. The most important lessons are to maintain decision-making for workloads involving a lot of data by offering proactive system layout policies. The established threshold detection and preemptive intervention techniques are part of a new paradigm of predictive system management that we have developed.

Chicken Swarm Optimization based QoS Web Service Composition in Cloud Computing Environment

2026-03-03T13:17:46+03:00

The increasing volume of data in the cloud computing environment has led to a situation where a single, abstract cloud web service has lost its ability to meet the complex requirements of various customers. Therefore, a service composition is becoming a necessity in the cloud computing environment. In this paper, Chicken Swarm Optimization (CSO) based QoS Web Service Composition in Cloud Computing Environment CSOQSC_CCE algorithm has been proposed. CSO is a metaheuristic algorithm that inspired from the hierarchy and foraging habits of the chicken swarm. The proposed CSOQSC_CCE algorithm is validated through using the Quality of Web Service (QWS) which is a real-world service dataset. The investigation results show that the proposed CSOQSC_CCE algorithm overcome the existing approaches in terms of response time and delay.

Lost in Translation? Detecting Emotional Drift in Bilingual Texts of Gilgamesh Using NLP

2026-02-25T15:25:13+03:00

The present work introduces the notion of "emotional drift" in literary translation, which refers to the diversion or transformation of emotional signals from source text to its translations. A bilingual analysis of the Arabic and English translations of The Epic of Gilgamesh serves as the focus of this paper, which, by utilizing recent expansions in Natural Language Processing (NLP) with an emphasis on sentiment analysis and emotion recognition, aims to quantify emotional content. The eventual objectives of this research are to develop a computational pipeline whereby emotional content can be extracted from and compared across bilingual literary corpora, to quantify emotional drift across varying versions of the Gilgamesh, and to contextualize these findings against the larger backdrop of translational research. The intersection of computational linguistics, translation theory, and digital literary analysis opens up a new avenue for machine-assisted literary scholarship and provides a replicable framework for inquiry into the issue of affective fidelity concerning translated literature.

MG Experimental Evaluation of Adaptive Multi-Agent AI Models for Detecting Stealthy Cyber Attacks in Dynamic Network Environments based on MADDPG

2026-02-10T23:00:23+03:00

Cybersecurity faces increasing challenges due to sophisticated and covert attacks targeting dynamic networks. This research aims to explore the effectiveness of adaptive multi-agent AI models in detecting these attacks using an algorithm MADDPG (Multi-Agent Deep Deterministic Policy Gradient) a dynamic network environment was designed to simulate natural data traffic and covert cyberattacks, where each agent monitors a specific part of the network and makes immediate decisions to detect or mitigate attacks .