Pub Date : 2026-03-01Epub Date: 2025-12-29DOI: 10.1016/j.eij.2025.100876
Dongliang Zhang , Lei Wang
Real-time visual image identification presents significant challenges due to noise, variations in illumination, and intricate backdrops, frequently resulting in misclassification and heightened processing costs. To mitigate these constraints, we offer a Fuzzy Dependency Model for Image Identification (FDM-II) that explicitly characterizes pixel interdependencies and executes adaptive feature selection. The approach incorporates fuzzification, fuzzy derivative optimization, and defuzzification to dynamically prioritize high-dependency features, minimize duplicate computation, and enhance classification robustness in uncertain settings. Utilizing the Open Images dataset, FDM-II attained 11.43% superior detection precision, 9.84% enhanced correlation rate, and 9.55% augmented classification accuracy relative to established RSS-based, TOPSIS-MADM, and fuzzy VHO methodologies, concurrently decreasing detection error and processing time by 8.77% and 10.06%, respectively. In contrast to conventional fixed-threshold or resource-intensive deep learning models, our methodology employs adaptive correlation-based refinement and dynamic feature ranking, facilitating scalable, low-latency, and reliable real-time performance appropriate for IoT and embedded applications.
{"title":"Quality-Aware Fuzzy-Logic-Based vertical handover decision method for dependable Real-Time visual image identification","authors":"Dongliang Zhang , Lei Wang","doi":"10.1016/j.eij.2025.100876","DOIUrl":"10.1016/j.eij.2025.100876","url":null,"abstract":"<div><div>Real-time visual image identification presents significant challenges due to noise, variations in illumination, and intricate backdrops, frequently resulting in misclassification and heightened processing costs. To mitigate these constraints, we offer a Fuzzy Dependency Model for Image Identification (FDM-II) that explicitly characterizes pixel interdependencies and executes adaptive feature selection. The approach incorporates fuzzification, fuzzy derivative optimization, and defuzzification to dynamically prioritize high-dependency features, minimize duplicate computation, and enhance classification robustness in uncertain settings. Utilizing the Open Images dataset, FDM-II attained 11.43% superior detection precision, 9.84% enhanced correlation rate, and 9.55% augmented classification accuracy relative to established RSS-based, TOPSIS-MADM, and fuzzy VHO methodologies, concurrently decreasing detection error and processing time by 8.77% and 10.06%, respectively. In contrast to conventional fixed-threshold or resource-intensive deep learning models, our methodology employs adaptive correlation-based refinement and dynamic feature ranking, facilitating scalable, low-latency, and reliable real-time performance appropriate for IoT and embedded applications.</div></div>","PeriodicalId":56010,"journal":{"name":"Egyptian Informatics Journal","volume":"33 ","pages":"Article 100876"},"PeriodicalIF":4.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145884769","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-02-14DOI: 10.1016/j.eij.2026.100911
Sitan Liu , Quanxi Feng , Wu Ai , Huazhou Chen , Bin Lin
Accurate commodity segmentation plays a crucial role in enhancing the competitiveness of sales enterprises in marketing. Currently, the retail industry widely employs cluster analysis and association rule algorithms for commodity segmentation and data mining.
The K-Means algorithm is widely used due to its simplicity, fast convergence, and suitability for large-scale datasets. However, traditional K-Means suffers from issues such as sensitivity to initial cluster centers, inability to handle mixed-type data, and ignoring relationships between attributes. While association rule mining effectively uncovers relationships between attributes, it is generally applied to categorical or discretized data and may generate an overly large set of candidate rules. To address these challenges, this paper proposes a novel clustering algorithm based on adaptive association rules, named AAP-KM.
The algorithm first uses adaptive association rules (AAP) to partition the dataset and obtain an initial division. It then calculates the initial cluster centers based on this partition, followed by the application of the K-Means algorithm for clustering. The main distinction of AAP-KM from traditional clustering methods is that it incorporates attribute relationships to determine more representative initial cluster centers. Additionally, the algorithm enhances its adaptability to different types of datasets by employing a secondary attribute transformation technique. To evaluate its effectiveness, numerical experiments are conducted on eight UCI datasets, with comparisons made against other improved K-Means algorithms. Experimental results demonstrate that AAP-KM exhibits significant performance advantages across multiple datasets. Finally, the AAP-KM algorithm is applied to the task of product segmentation.
{"title":"Research on K-Means algorithm based on adaptive association rules and its application in commodity segmentation","authors":"Sitan Liu , Quanxi Feng , Wu Ai , Huazhou Chen , Bin Lin","doi":"10.1016/j.eij.2026.100911","DOIUrl":"10.1016/j.eij.2026.100911","url":null,"abstract":"<div><div>Accurate commodity segmentation plays a crucial role in enhancing the competitiveness of sales enterprises in marketing. Currently, the retail industry widely employs cluster analysis and association rule algorithms for commodity segmentation and data mining.</div><div>The K-Means algorithm is widely used due to its simplicity, fast convergence, and suitability for large-scale datasets. However, traditional K-Means suffers from issues such as sensitivity to initial cluster centers, inability to handle mixed-type data, and ignoring relationships between attributes. While association rule mining effectively uncovers relationships between attributes, it is generally applied to categorical or discretized data and may generate an overly large set of candidate rules. To address these challenges, this paper proposes a novel clustering algorithm based on adaptive association rules, named AAP-KM.</div><div>The algorithm first uses adaptive association rules (AAP) to partition the dataset and obtain an initial division. It then calculates the initial cluster centers based on this partition, followed by the application of the K-Means algorithm for clustering. The main distinction of AAP-KM from traditional clustering methods is that it incorporates attribute relationships to determine more representative initial cluster centers. Additionally, the algorithm enhances its adaptability to different types of datasets by employing a secondary attribute transformation technique. To evaluate its effectiveness, numerical experiments are conducted on eight UCI datasets, with comparisons made against other improved K-Means algorithms. Experimental results demonstrate that AAP-KM exhibits significant performance advantages across multiple datasets. Finally, the AAP-KM algorithm is applied to the task of product segmentation.</div></div>","PeriodicalId":56010,"journal":{"name":"Egyptian Informatics Journal","volume":"33 ","pages":"Article 100911"},"PeriodicalIF":4.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146187900","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2025-12-27DOI: 10.1016/j.eij.2025.100873
Nevzat Olgun
In this study, a novel method based on Variational Mode Decomposition (VMD) is proposed for lie detection from EEG signals (EEGs). The study was conducted using the LieWaves database, and analyses were performed on 5 −channel EEGs obtained from 27 subjects. The EEGs collected from the subjects during truthful and lying situations were divided into 2-second segments based on the moments when visual stimuli were presented, and a total of 1350 EEG signals were obtained. For lie detection, 3 channels were selected, and EEG signals were processed using the VMD technique and time domain features were extracted from each mode. Extra Trees, Random Forest, K-Nearest Neighbors and Support Vector Machine classification models were used to classify the data. As a result of the tests, the Extra Trees model achieved the highest performance, reaching 100% classification accuracy. The other classification models achieved 99.93%, 99.48% and 64.22% classification accuracy, respectively. These results show that the VMD-based method provides an effective and efficient solution for EEG-based lie detection and it is suitable for real-time applications on portable EEG devices. Moreover, the proposed method is more advantageous than the complex approaches in the literature with its low number of channels and low processing time. The results show that this method has great potential for future studies and applications in the detection of deception.
{"title":"A novel method based on variational mode decomposition for lie detection","authors":"Nevzat Olgun","doi":"10.1016/j.eij.2025.100873","DOIUrl":"10.1016/j.eij.2025.100873","url":null,"abstract":"<div><div>In this study, a novel method based on Variational Mode Decomposition (VMD) is proposed for lie detection from EEG signals (EEGs). The study was conducted using the LieWaves database, and analyses were performed on 5 −channel EEGs obtained from 27 subjects. The EEGs collected from the subjects during truthful and lying situations were divided into 2-second segments based on the moments when visual stimuli were presented, and a total of 1350 EEG signals were obtained. For lie detection, 3 channels were selected, and EEG signals were processed using the VMD technique and time domain features were extracted from each mode. Extra Trees, Random Forest, K-Nearest Neighbors and Support Vector Machine classification models were used to classify the data. As a result of the tests, the Extra Trees model achieved the highest performance<strong>,</strong> reaching 100% classification accuracy. The other classification models achieved 99.93%, 99.48% and 64.22% classification accuracy, respectively. These results show that the VMD-based method provides an effective and efficient solution for EEG-based lie detection and it is suitable for real-time applications on portable EEG devices. Moreover, the proposed method is more advantageous than the complex approaches in the literature with its low number of channels and low processing time. The results show that this method has great potential for future studies and applications in the detection of deception.</div></div>","PeriodicalId":56010,"journal":{"name":"Egyptian Informatics Journal","volume":"33 ","pages":"Article 100873"},"PeriodicalIF":4.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145841260","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-01-27DOI: 10.1016/j.eij.2026.100890
Jian Zheng , Shengye Wang , Huyong Yan , Haichao Sun
This work proposes a kernel amplification method with non-stationary characteristics for binary classification of non-noisy imbalanced datasets. Our methodology features two key innovations, including that a derived non-stationary kernel construction enables adaptive exploration of minority class regions, and a Riemannian metric–guided kernel amplification mechanism effectively induces minority class migration in feature space, tightening the spatial distance inner minority class instances. Experimental validation across ten UCI benchmark datasets with class imbalance demonstrate the superior performance of our proposed method. The method achieves statistically significant superiority over all six baseline approaches on five highly imbalanced datasets (with imbalance ratios (IR) > 10:1), notably achieving 0.883 F1-score on datasets with 40.22:1 imbalance ratio and 0.800 sensitivity to the minority class. Furthermore, our approach maintains competitive advantages on the remaining five moderately imbalanced datasets (IR < 10:1), outperforming a subset of the baseline methods across all evaluation metrics. Furthermore, the kernel amplification mechanism boosts the sensitivity to perception minority classes by a maximum 6.35-fold enhancement on highly imbalanced datasets, and by a maximum 2.17-fold enhancement on moderately imbalanced datasets. The derived amplification factor exhibits dimension-dependent characteristics, showing independence from both sample size and imbalanced ratio——a critical advantage for high-dimensional imbalanced classification.
{"title":"Binary classification for imbalanced datasets using a novel metric method","authors":"Jian Zheng , Shengye Wang , Huyong Yan , Haichao Sun","doi":"10.1016/j.eij.2026.100890","DOIUrl":"10.1016/j.eij.2026.100890","url":null,"abstract":"<div><div>This work proposes a kernel amplification method with non-stationary characteristics for binary classification of non-noisy imbalanced datasets. Our methodology features two key innovations, including that a derived non-stationary kernel construction enables adaptive exploration of minority class regions, and a Riemannian metric–guided kernel amplification mechanism effectively induces minority class migration in feature space, tightening the spatial distance inner minority class instances. Experimental validation across ten UCI benchmark datasets with class imbalance demonstrate the superior performance of our proposed method. The method achieves statistically significant superiority over all six baseline approaches on five highly imbalanced datasets (with imbalance ratios (IR) > 10:1), notably achieving 0.883 F1-score on datasets with 40.22:1 imbalance ratio and 0.800 sensitivity to the minority class. Furthermore, our approach maintains competitive advantages on the remaining five moderately imbalanced datasets (IR < 10:1), outperforming a subset of the baseline methods across all evaluation metrics. Furthermore, the kernel amplification mechanism boosts the sensitivity to perception minority classes by a maximum 6.35-fold enhancement on highly imbalanced datasets, and by a maximum 2.17-fold enhancement on moderately imbalanced datasets. The derived amplification factor exhibits dimension-dependent characteristics, showing independence from both sample size and imbalanced ratio——a critical advantage for high-dimensional imbalanced classification.</div></div>","PeriodicalId":56010,"journal":{"name":"Egyptian Informatics Journal","volume":"33 ","pages":"Article 100890"},"PeriodicalIF":4.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146078375","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The rapid integration of photovoltaic and wind-based distributed energy resources (DERs) into radial distribution networks has introduced operational challenges such as voltage instability, increased losses, and unpredictable system behaviour under renewable variability. These issues require optimization frameworks that are both computationally efficient and capable of modelling uncertainty. This paper presents a Machine Learning-Enhanced Cheetah Optimizer (ML-EChOA) that integrates time–frequency voltage analysis with surrogate-assisted metaheuristic search to achieve fast and accurate techno-economic DER allocation. Voltage time series are transformed into spectrograms and scalograms, from which Local Binary Pattern features are extracted to capture transient behaviour. A Gradient Boosting surrogate is trained on these features to approximate power-flow outcomes, enabling the optimizer to evaluate candidate solutions with minimal computational overhead. Deterministic and probabilistic scenarios — generated through LSTM-based forecasting of solar, wind, and load profiles — ensure that the optimization remains robust under uncertainty. The proposed approach produces substantially improved voltage quality, reduced losses, and enhanced economic performance while converging faster than conventional metaheuristics. These results illustrate the potential of ML-EChOA as a scalable, intelligent, and uncertainty-aware optimization tool for renewable integration and future smart distribution networks.
{"title":"Machine learning and time–frequency feature framework for optimal DER planning in radial networks","authors":"Sangeeta DebBarman , Kumari Namrata , Manoj Gupta , Pannee Suanpang , Aziz Nanthaamornphong","doi":"10.1016/j.eij.2026.100928","DOIUrl":"10.1016/j.eij.2026.100928","url":null,"abstract":"<div><div>The rapid integration of photovoltaic and wind-based distributed energy resources (DERs) into radial distribution networks has introduced operational challenges such as voltage instability, increased losses, and unpredictable system behaviour under renewable variability. These issues require optimization frameworks that are both computationally efficient and capable of modelling uncertainty. This paper presents a Machine Learning-Enhanced Cheetah Optimizer (ML-EChOA) that integrates time–frequency voltage analysis with surrogate-assisted metaheuristic search to achieve fast and accurate techno-economic DER allocation. Voltage time series are transformed into spectrograms and scalograms, from which Local Binary Pattern features are extracted to capture transient behaviour. A Gradient Boosting surrogate is trained on these features to approximate power-flow outcomes, enabling the optimizer to evaluate candidate solutions with minimal computational overhead. Deterministic and probabilistic scenarios — generated through LSTM-based forecasting of solar, wind, and load profiles — ensure that the optimization remains robust under uncertainty. The proposed approach produces substantially improved voltage quality, reduced losses, and enhanced economic performance while converging faster than conventional metaheuristics. These results illustrate the potential of ML-EChOA as a scalable, intelligent, and uncertainty-aware optimization tool for renewable integration and future smart distribution networks.</div></div>","PeriodicalId":56010,"journal":{"name":"Egyptian Informatics Journal","volume":"33 ","pages":"Article 100928"},"PeriodicalIF":4.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147396707","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-01-23DOI: 10.1016/j.eij.2026.100887
Rasha M. Abd El-Aziz, Alanazi Rayan
Stroke is a leading cause of global mortality and long-term disability, emphasizing the urgent need for predictive models that are accurate, interpretable, and equitable to support precision medicine. Conventional risk assessment methods often rely on a limited set of clinical indicators and ignore subgroup-specific patterns, which reduces predictive performance and can bias outcomes against underrepresented populations. To address these challenges, this study proposes ASTab-Stroke (Adaptive Stratified TabNet for Stroke Prediction), a deep learning framework integrating Adaptive Stratified Sampling (ASS) with TabNet’s sequential attention mechanism. ASS dynamically reweights patient strata based on their contribution to prediction errors, ensuring fair representation of minority and high-risk subgroups without introducing synthetic data. TabNet’s sequential attention provides step-wise feature attribution, enabling clinicians to interpret the influence of predictors such as age, hypertension, heart disease, glucose level, BMI, and lifestyle factors on stroke risk. The framework was implemented in Python 3.10 and evaluated using the Stroke Prediction Dataset, which includes diverse demographic, clinical, and lifestyle variables. ASTab-Stroke achieved 98% accuracy, 0.998 AUC, 0.97 F1-score, 0.99 recall, and 0.98 precision, outperforming existing baselines by approximately 3% in accuracy while demonstrating improved sensitivity and fairness across clinically significant subgroups. The age and comorbidity features proved to be critical in ablation studies and work on cross-validation showed strong generalization. This framework is a clinically interpretable, scalable, and ethically rationalized method of stroke risk prediction, which gives dependable information to support clinical decision-making with data. The flexibility of it implies that it has a wide potential to be used in other fields of precision medicine, where interpretability and subgroup fairness are crucial in promoting equitable and informed patient care.
中风是全球死亡和长期残疾的主要原因，因此迫切需要准确、可解释和公平的预测模型来支持精准医学。传统的风险评估方法往往依赖于一组有限的临床指标，忽略了亚组特定模式，这降低了预测效果，并可能使结果对代表性不足的人群产生偏差。为了解决这些挑战，本研究提出了ASTab-Stroke (Adaptive Stratified TabNet for Stroke Prediction)，这是一个将自适应分层采样（ASS）与TabNet的顺序注意机制相结合的深度学习框架。ASS根据患者阶层对预测误差的贡献动态地重新加权，确保在不引入合成数据的情况下公平地代表少数群体和高风险亚群。TabNet的顺序关注提供了阶梯式特征归因，使临床医生能够解释诸如年龄、高血压、心脏病、血糖水平、BMI和生活方式等预测因素对中风风险的影响。该框架在Python 3.10中实现，并使用中风预测数据集进行评估，该数据集包括各种人口统计、临床和生活方式变量。ASTab-Stroke的准确度为98%，AUC为0.998，f1评分为0.97，召回率为0.99，精密度为0.98，准确度比现有基线提高了约3%，同时在临床显著亚组中表现出更高的敏感性和公平性。年龄和合并症特征在消融研究中被证明是至关重要的，交叉验证的工作显示出很强的通用性。该框架是一种临床可解释、可扩展、伦理合理的脑卒中风险预测方法，为临床决策提供可靠的数据支持。它的灵活性意味着它在其他精准医学领域具有广泛的应用潜力，在这些领域，可解释性和亚组公平性对于促进公平和知情的患者护理至关重要。
{"title":"Adaptive sampling enhanced deep learning framework for accurate interpretable stroke risk prediction","authors":"Rasha M. Abd El-Aziz, Alanazi Rayan","doi":"10.1016/j.eij.2026.100887","DOIUrl":"10.1016/j.eij.2026.100887","url":null,"abstract":"<div><div>Stroke is a leading cause of global mortality and long-term disability, emphasizing the urgent need for predictive models that are accurate, interpretable, and equitable to support precision medicine. Conventional risk assessment methods often rely on a limited set of clinical indicators and ignore subgroup-specific patterns, which reduces predictive performance and can bias outcomes against underrepresented populations. To address these challenges, this study proposes ASTab-Stroke (Adaptive Stratified TabNet for Stroke Prediction), a deep learning framework integrating Adaptive Stratified Sampling (ASS) with TabNet’s sequential attention mechanism. ASS dynamically reweights patient strata based on their contribution to prediction errors, ensuring fair representation of minority and high-risk subgroups without introducing synthetic data. TabNet’s sequential attention provides step-wise feature attribution, enabling clinicians to interpret the influence of predictors such as age, hypertension, heart disease, glucose level, BMI, and lifestyle factors on stroke risk. The framework was implemented in Python 3.10 and evaluated using the Stroke Prediction Dataset, which includes diverse demographic, clinical, and lifestyle variables. ASTab-Stroke achieved 98% accuracy, 0.998 AUC, 0.97 F1-score, 0.99 recall, and 0.98 precision, outperforming existing baselines by approximately 3% in accuracy while demonstrating improved sensitivity and fairness across clinically significant subgroups. The age and comorbidity features proved to be critical in ablation studies and work on cross-validation showed strong generalization. This framework is a clinically interpretable, scalable, and ethically rationalized method of stroke risk prediction, which gives dependable information to support clinical decision-making with data. The flexibility of it implies that it has a wide potential to be used in other fields of precision medicine, where interpretability and subgroup fairness are crucial in promoting equitable and informed patient care.</div></div>","PeriodicalId":56010,"journal":{"name":"Egyptian Informatics Journal","volume":"33 ","pages":"Article 100887"},"PeriodicalIF":4.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146037846","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-02-11DOI: 10.1016/j.eij.2026.100904
Ahmed Soliman, Khalid M. Amin, Noura A. Semary, Hayam Mousa
Federated Learning (FL) enables collaborative model training without sharing raw data but remains vulnerable to backdoor attacks, particularly under high adversarial ratios and non-IID data distributions. Existing defenses often rely on clean public datasets or strong threat assumptions, limiting real-world applicability. We propose SHIELD-FL (Secure Hybrid Inspection for Enhanced Learning Defense in Federated Learning), a scalable, attack-agnostic framework that operates without external data. SHIELD-FL integrates: (1) Gradient Trust Indexing, which dynamically scores client reliability via adversarial perturbation sensitivity; (2) Adaptive Clustering using HDBSCAN in parameter space to isolate benign clients; and (3) Robust Knowledge Distillation with temperature-scaled soft labels and stochastic weight averaging. Extensive evaluations on CIFAR-10, EMNIST, and Fashion-MNIST under five adaptive backdoor attacks show SHIELD-FL achieves up to 92.5% main-task accuracy while reducing attack success rates to 3.6%, even with 60% malicious clients. It outperforms data-dependent defenses like FLTrust, maintains low communication overhead, and runs 3–4 faster than ensemble methods. SHIELD-FL is especially suitable for privacy-sensitive, resource-constrained environments including emerging MENA region applications.
{"title":"SHIELD-FL: Scalable backdoor defense in federated learning via gradient trust and data-free distillation under non-IID data","authors":"Ahmed Soliman, Khalid M. Amin, Noura A. Semary, Hayam Mousa","doi":"10.1016/j.eij.2026.100904","DOIUrl":"10.1016/j.eij.2026.100904","url":null,"abstract":"<div><div>Federated Learning (FL) enables collaborative model training without sharing raw data but remains vulnerable to backdoor attacks, particularly under high adversarial ratios and non-IID data distributions. Existing defenses often rely on clean public datasets or strong threat assumptions, limiting real-world applicability. We propose <strong>SHIELD-FL</strong> (Secure Hybrid Inspection for Enhanced Learning Defense in Federated Learning), a scalable, attack-agnostic framework that operates without external data. SHIELD-FL integrates: (1) <em>Gradient Trust Indexing</em>, which dynamically scores client reliability via adversarial perturbation sensitivity; (2) <em>Adaptive Clustering</em> using HDBSCAN in parameter space to isolate benign clients; and (3) <em>Robust Knowledge Distillation</em> with temperature-scaled soft labels and stochastic weight averaging. Extensive evaluations on CIFAR-10, EMNIST, and Fashion-MNIST under five adaptive backdoor attacks show SHIELD-FL achieves up to 92.5% main-task accuracy while reducing attack success rates to 3.6%, even with 60% malicious clients. It outperforms data-dependent defenses like FLTrust, maintains low communication overhead, and runs 3–4<span><math><mo>×</mo></math></span> faster than ensemble methods. SHIELD-FL is especially suitable for privacy-sensitive, resource-constrained environments including emerging MENA region applications.</div></div>","PeriodicalId":56010,"journal":{"name":"Egyptian Informatics Journal","volume":"33 ","pages":"Article 100904"},"PeriodicalIF":4.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146187898","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-02-12DOI: 10.1016/j.eij.2026.100918
Wei Zhang , Yuanbin Mo
Accurate parameter identification for photovoltaic (PV) cells is essential for the evaluation, control, and improvement of PV systems. To address this, various metaheuristic algorithms have been widely employed. Most of the methods and techniques found in the existing literature rely on Root Mean Square Error (RMSE) values to validate the performance of the approaches employed. The objective of this study is to propose a modified RMSE calculation formula, providing a more precise evaluation metric for optimization algorithms used to extract PV model parameters. The RMSE values of the RTC France solar cell is first calculated using 30 distinct algorithms. Next, a multi-strategy improved partial reinforcement optimization algorithm (IAF-PRO) is developed, which incorporates an improved nonlinear selection rate and adaptive strong and weak stimulation factors. This modification enhances the algorithm’s convergence speed and precision. Subsequently, the collaboration of mirror reflection boundary handling mechanism and fast random opposition-based learning synergistically promotes population diversity and assists the algorithm in avoiding local optima. Finally, the suggested IAF-PRO is employed to estimate parameters for various PV models, including single diode, double diode, and PV module. Experimental results show that IAF-PRO consistently achieves the lowest RMSE values across all five PV models compared to nine state-of-the-art algorithms, including PSO and GWO, establishing its efficacy for PV parameter estimation.
{"title":"Multi-strategy improved partial reinforcement optimization algorithm for accurate photovoltaic parameter extraction","authors":"Wei Zhang , Yuanbin Mo","doi":"10.1016/j.eij.2026.100918","DOIUrl":"10.1016/j.eij.2026.100918","url":null,"abstract":"<div><div>Accurate parameter identification for photovoltaic (PV) cells is essential for the evaluation, control, and improvement of PV systems. To address this, various metaheuristic algorithms have been widely employed. Most of the methods and techniques found in the existing literature rely on Root Mean Square Error (RMSE) values to validate the performance of the approaches employed. The objective of this study is to propose a modified RMSE calculation formula, providing a more precise evaluation metric for optimization algorithms used to extract PV model parameters. The RMSE values of the RTC France solar cell is first calculated using 30 distinct algorithms. Next, a multi-strategy improved partial reinforcement optimization algorithm (IAF-PRO) is developed, which incorporates an improved nonlinear selection rate and adaptive strong and weak stimulation factors. This modification enhances the algorithm’s convergence speed and precision. Subsequently, the collaboration of mirror reflection boundary handling mechanism and fast random opposition-based learning synergistically promotes population diversity and assists the algorithm in avoiding local optima. Finally, the suggested IAF-PRO is employed to estimate parameters for various PV models, including single diode, double diode, and PV module. Experimental results show that IAF-PRO consistently achieves the lowest RMSE values across all five PV models compared to nine state-of-the-art algorithms, including PSO and GWO, establishing its efficacy for PV parameter estimation.</div></div>","PeriodicalId":56010,"journal":{"name":"Egyptian Informatics Journal","volume":"33 ","pages":"Article 100918"},"PeriodicalIF":4.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146187897","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-02-13DOI: 10.1016/j.eij.2026.100898
Xiancheng Chen
In recent years, the complexity of the global supply chain has increased, accompanied by heightened risks, uncertainties, and catastrophic events. Supply chain resilience has become an important factor, particularly for dynamic industries such as live-streaming e-commerce, where enterprises must adapt to sudden market shifts and unpredictable disruptions. However, traditional supply chain models lack efficient mechanisms to integrate decentralized data sources, leading to inefficient supplier selection, resource allocation, and real-time decision-making. The research proposes a Service-Specific Data Management Model (S2DM2), a novel framework integrating reverse engineering and federated learning to enhance supply chain resilience through optimized data processing. S2DM2 introduces a reverse data handling mechanism to mitigate timeline-based data segregation problems, thus improving central service management and supplier evaluation processes. By analyzing historical and real-time supply, production, and delivery data, S2DM2 optimizes supplier selection, route planning, and resource distribution, making it significantly suitable for fast-evolving industries like live streaming e-commerce. Furthermore, its classified federated learning architecture provides a decentralized way of data sharing while preserving privacy and efficiency in decision-making. Experimental tests indicate that S2DM2 improves data mapping accuracy by 8.89%, computation time by 9.92%, and map failure rates by 8.76% compared to existing supply chain management models. These increments form the basis of scalable, service-oriented supply chain operations, and hence, S2DM2 is implementable for SMEs while navigating dynamic digitalized markets.
{"title":"Service oriented supply chain optimization in e-commerce with federated learning and reverse data handling","authors":"Xiancheng Chen","doi":"10.1016/j.eij.2026.100898","DOIUrl":"10.1016/j.eij.2026.100898","url":null,"abstract":"<div><div>In recent years, the complexity of the global supply chain has increased, accompanied by heightened risks, uncertainties, and catastrophic events. Supply chain resilience has become an important factor, particularly for dynamic industries such as live-streaming e-commerce, where enterprises must adapt to sudden market shifts and unpredictable disruptions. However, traditional supply chain models lack efficient mechanisms to integrate decentralized data sources, leading to inefficient supplier selection, resource allocation, and real-time decision-making. The research proposes a Service-Specific Data Management Model (S2DM2), a novel framework integrating reverse engineering and federated learning to enhance supply chain resilience through optimized data processing. S2DM2 introduces a reverse data handling mechanism to mitigate timeline-based data segregation problems, thus improving central service management and supplier evaluation processes. By analyzing historical and real-time supply, production, and delivery data, S2DM2 optimizes supplier selection, route planning, and resource distribution, making it significantly suitable for fast-evolving industries like live streaming e-commerce. Furthermore, its classified federated learning architecture provides a decentralized way of data sharing while preserving privacy and efficiency in decision-making. Experimental tests indicate that S2DM2 improves data mapping accuracy by 8.89%, computation time by 9.92%, and map failure rates by 8.76% compared to existing supply chain management models. These increments form the basis of scalable, service-oriented supply chain operations, and hence, S2DM2 is implementable for SMEs while navigating dynamic digitalized markets.</div></div>","PeriodicalId":56010,"journal":{"name":"Egyptian Informatics Journal","volume":"33 ","pages":"Article 100898"},"PeriodicalIF":4.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146187896","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-02-06DOI: 10.1016/j.eij.2026.100907
Zisheng Li , Xiaoping Xiao , Honghao Fu , Tengfei Jiang , Lin Jing , Wen Xiong
Tool wear monitoring is essential for intelligent manufacturing. Although multi-source domain generalization methods do not rely on target-domain data, their practical deployment is still constrained because they require labeled samples from multiple source domains. This limitation mainly stems from the fact that labeled data across operating conditions are often difficult to obtain and scarce in industrial scenarios. Moreover, under the single-source domain generalization setting, conventional data generation strategies typically emphasize sample diversity while neglecting the fidelity and realism of the generated samples, which can in turn undermine model reliability and generalization. To address these challenges, this paper proposes a Single-source domain tool wear prediction based on generated feature generalization. The proposed approach first performs denoising preprocessing on raw vibration signals and then employs an AdaIN-CNN to generate samples that are semantically consistent with the source domain while exhibiting controlled distributional differences, thereby expanding the coverage of single-source-domain data. During the generator training stage, under the constraint of the mean absolute error (DDL), mutual information (MI) and maximum mean discrepancy (MMD) are jointly introduced to optimize the relationship between source-domain and generated-domain features, improving feature diversity and cross-condition robustness while maintaining generation fidelity. During the prediction model training stage, the mutual information between source and generated features is maximized, forming an adversarial interplay with the generator, which further enhances the generalization capability of the predictive model under unseen operating conditions. The proposed method is validated on both the public NASA dataset and the experimental dataset collected from a self-built platform. Experimental results show that the proposed method achieves RMSE/ scores of 0.0992/0.7226 on the NASA benchmark dataset and 0.3177/0.7630 on the self-built dataset, outperforming other baseline methods.
{"title":"Single-source domain tool wear prediction based on generated feature generalization","authors":"Zisheng Li , Xiaoping Xiao , Honghao Fu , Tengfei Jiang , Lin Jing , Wen Xiong","doi":"10.1016/j.eij.2026.100907","DOIUrl":"10.1016/j.eij.2026.100907","url":null,"abstract":"<div><div>Tool wear monitoring is essential for intelligent manufacturing. Although multi-source domain generalization methods do not rely on target-domain data, their practical deployment is still constrained because they require labeled samples from multiple source domains. This limitation mainly stems from the fact that labeled data across operating conditions are often difficult to obtain and scarce in industrial scenarios. Moreover, under the single-source domain generalization setting, conventional data generation strategies typically emphasize sample diversity while neglecting the fidelity and realism of the generated samples, which can in turn undermine model reliability and generalization. To address these challenges, this paper proposes a Single-source domain tool wear prediction based on generated feature generalization. The proposed approach first performs denoising preprocessing on raw vibration signals and then employs an AdaIN-CNN to generate samples that are semantically consistent with the source domain while exhibiting controlled distributional differences, thereby expanding the coverage of single-source-domain data. During the generator training stage, under the constraint of the mean absolute error (DDL), mutual information (MI) and maximum mean discrepancy (MMD) are jointly introduced to optimize the relationship between source-domain and generated-domain features, improving feature diversity and cross-condition robustness while maintaining generation fidelity. During the prediction model training stage, the mutual information between source and generated features is maximized, forming an adversarial interplay with the generator, which further enhances the generalization capability of the predictive model under unseen operating conditions. The proposed method is validated on both the public NASA dataset and the experimental dataset collected from a self-built platform. Experimental results show that the proposed method achieves RMSE/<span><math><msup><mrow><mi>R</mi></mrow><mrow><mn>2</mn></mrow></msup></math></span> scores of 0.0992/0.7226 on the NASA benchmark dataset and 0.3177/0.7630 on the self-built dataset, outperforming other baseline methods.</div></div>","PeriodicalId":56010,"journal":{"name":"Egyptian Informatics Journal","volume":"33 ","pages":"Article 100907"},"PeriodicalIF":4.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146188598","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号