ABSTRACT. The O-RAN is one of the emerging technologies that can help reduce handoff failures. O-RAN, built on open and interoperable RAN nodes, enables the integration of AI through modular xApps within the Near-RT RIC. This paper proposes a system model for load-balancing handover using three distinct xApps. These xApps are divided into three sequential nodes, where each node is responsible for predicting an outcome that serves as the input for the next xApp. The first node is the Anomaly Detection (AD) xApp, which predicts anomalous traffic patterns. This output is then utilized by the second node for Load Balancing xApp, allowing the system to forecast the performance of both the serving and neighboring cells for the next five steps. Finally, these parameters are provided to the Handover xApp, which makes the definitive decision to hand off the User Equipment (UE) from one cell to another. In this paper, machine learning techniques and fine-tuning were applied across all three nodes. The AI models were primarily evaluated based on accuracy and training time. Training time is particularly critical in this context, as xApps operate in a closed-loop environment with strict latency requirements ranging from 10 ms to 1 second. The second proposed model based on the CatBoost which will outperform XGBoost by improving the overall feature extraction accuracy by 0.66% as well as improving the training time and reducing it by 95.04%. this will reflect a great deployment from the scenario from computational efficiency perspective.

ABSTRACT. The rapid expansion of the Internet of Things (IoT) has intensified the need for sustainable, battery‑free communication solutions. This paper presents a comprehensive Simulink model of a modified Passive LiFi system that combines Visible Light Communication (VLC) for optical downlink with RF backscatter for ultra‑low‑power uplink transmission. To overcome limitations in standard Simulink libraries, custom behavioral models are developed for key components, including a solar cell equivalent circuit, a boost converter replacing the switching regulator for stable high‑frequency modulation, and an optocoupler‑based gate driver. The model encompasses the complete communication chain: LiFi transmitter with chirp‑modulated OOK, tag‑side signal conditioning (HPF, comparator), and CSS demodulation at the receiver. System performance was evaluated in terms of BER, SNR, CDF analysis confirms close statistical agreement with experimental Passive LiFi results. The framework provides a flexible, reproducible platform for early‑stage design exploration of hybrid LiFi–RF backscatter systems without costly hardware prototyping.

ABSTRACT. Predicting customer attrition is no longer just a matter of classification accuracy; it has become a challenge to understand the inherent uncertainty in human behavior. This paper proposes a robust churn prediction framework based on Natural Gradient Boosting (NGBoost). The experiments were conducted on a telecom dataset which contains 7,043 customers with 21 features, the current approach integrates derived features with hybrid resampling strategy combining Synthetic Minority Oversampling Technique (SMOTE) and random under-sampling to mitigate class imbalance. The results showed that NGBoost achieved an accuracy of 82.7% and an AUC of 91.6%. In addition to classification, the proposed model provides well-calibrated probability scores, achieving a Brier score of 0.1211. Based on confidence levels, firms can define and prioritize intervention strategies for effective decision-making in the telecommunications sector.

ABSTRACT. Complex nonlinear and high-dimensional optimization problems are solved using metaheuristic optimization algorithms. Nevertheless, there is still a challenge of keeping an effective balance between exploration and exploitation. In this paper, a hybrid metaheuristic algorithm is proposed; it is named Dual-Strategy Walrus-Gazelle Optimizer (DEWGO). The suggested approach involves the exploration ability of the Gazelle Optimization Algorithm (GOA) together with the exploitation approach of the Walrus Optimizer (WO). A dynamically controlled switching parameter is an adaptive iteration-based parameter that dynamically controls the switching between exploration and exploitation in the search process. DEWGO is tested with the CEC2017, CEC2022 suits, and six real-world engineering problems and its performance metrics are compared to ten popular optimization algorithms. The experimental findings indicate that DEWGO performs even better, and achieves the best overall rank on most benchmark functions. The statistical analysis based on the Wilcoxon signed-rank test proves the significance of the obtained improvements.

ABSTRACT. Off-chain storage solutions like the InterPlanetary File System (IPFS) are vital for overcoming blockchain storage limitations. While IPFS Clusters ensure data availability, traditional replication-based models introduce high resource overhead and scalability bottlenecks. The Verifiable Decentralized IPFS Cluster (VDIC) addresses data integrity but remains limited by inefficient storage utilization. We propose EC-VDIC, an adaptive framework that integrates Reed-Solomon erasure coding to optimize verifiable storage. Using an adaptive gateway, EC-VDIC dynamically applies replication to small metadata and erasure coding to large data blocks. Evaluated across 3 to 20 nodes, EC-VDIC achieved a 62.05% reduction in physical storage compared to standard VDIC. Additionally, while replication-based models showed a double latency spike when scaling from 10 to 20 nodes, EC-VDIC maintained a small 12.3% increase. Our results confirm EC-VDIC as a scalable, storage-efficient, and verifiable solution for decentralized data management.

ABSTRACT. Reconfigurable Intelligent Surfaces (RIS) and Intelligent Reflecting Surfaces (IRS) are two prominent concepts in modern wireless communications. These surfaces are primarily used to enhance wireless channel quality, particularly in non-line-of-sight (NLOS) scenarios. In most applications, they are designed to introduce non-specular reflection paths between transmitter and receiver. However, other functionalities include multi-beam reflection, polarization-dependent reflection, and frequency-dependent reflection. Moreover, intelligent surfaces can also be transmissive or guiding rather than purely reflective. Additionally, active surfaces can provide power gain or even frequency conversion. These properties may suffice for static scenarios where reconfigurability is not essential. In contrast, applications such as vehicular communication systems require the surfaces to be reconfigurable. Therefore, the more general term Intelligent Surfaces (IS) can encompass reflecting, transmitting, guiding, reconfigurable, static, passive, or active surfaces, depending on application requirements. The implementation of such surfaces relies on designing artificial surfaces composed of arrays of sub-wavelength elements. For reconfigurable surfaces, control and processing systems are also necessary. This talk will introduce the fundamental theory for designing various configurations of intelligent surfaces, their implementation methods, and their applications in wireless communication systems. The focus is on the electromagnetic analysis and practical realization of these surfaces.

ABSTRACT. This paper offers a new lantern-shaped microstrip patch antenna considered for high-performance terahertz (THz) communications. The single-element antenna is made from a 10 μm flexible polyimide substrate. It has a rhombic-shaped patch with an elliptical handle and a candle-shaped slot to improve impedance matching and bandwidth. The optimized design achieves a wide operating range from 1.6 to 1.83 THz, with a gain of 7.5 dBi and 91% radiation efficiency. The design is also expanded to a MIMO configuration for better performance. A tree-shaped decoupling surface is introduced to lessen mutual coupling among elements. Machine learning techniques optimize the decoupling surface length, achieving high prediction accuracy (R² = 0.9908) and a minimum S21 of −57.72 dB. The planned MIMO antenna shows excellent diversity recital, with ECC < 10⁻³, a diversity gain of 9.998 dB, a peak gain of 9.5 dBi, and efficiency up to 99.6%.

ABSTRACT. In this paper, a new and flexible flower-shaped microstrip patch antenna is proposed, which is suitable for the terahertz (THz) band and is intended for the development of short-range communication in the upcoming 6G networks. The antenna is carefully designed using a polyimide substrate with a thickness of 10 μm and a monolayer graphene patch. The flower-shaped antenna is optimized using a step-by-step evolution algorithm. To overcome the gain limitation of THz antennas, a flower-shaped metasurface (MS) is used, which is located at an optimized distance of 200 μm. Simulation results reveal that the proposed antenna is resonant at a deep frequency of 0.516 THz with a reflection coefficient less than -30 dB. The proposed flower-shaped antenna with the MS demonstrates improved results, with the peak realized gain increasing from 6.5 dBi to 8.1 dBi and the radiation efficiency increasing to 80%. Moreover, the surface currents and radiation patterns confirm the effectiveness of the proposed MS in suppressing the surface waves and providing improved directivity.

ABSTRACT. In this paper, a novel symmetric nested square cross-stub terahertz metamaterial absorber is carried out for multi-band sensing purposes. The unit cell comprises a designed metallic resonator in the shape of nested squares with cross-stub connections in an orthogonal manner over a polyimide substrate with a metallic ground plane at the back side. The absorber shows four clear absorption peaks above 95% at 0.75 THz, 1.30 THz, 1.50 THz, and 2.70 THz under normal incidence for both TE and TM polarizations. The transmission coefficient is found to be nearly zero due to the presence of the metallic ground plane. The presence of deep minima in the reflection curves indicates effective impedance matching at the resonant frequencies. Due to the four-fold rotation symmetry of the structure, there is an excellent polarization insensitivity of the proposed SNSCS absorber. The absorber also shows good angular stability with high absorption values above 65-90% at the main resonant frequencies for oblique incidence angles up to 60° in both TE and TM modes. At a fixed large value of the incidence angle θ = 60°, the response is found to be nearly invariant over the range of azimuthal angles from φ = 10° to 60°, indicating the strong azimuthal robustness of the proposed SNSCS absorber. The multi-resonant frequency response is due to the coupled electric and magnetic dipole modes of the outer and inner nested structures of the SNSCS absorber, allowing for selective tuning of individual resonant frequencies. The proposed SNSCS absorber can be used in multi-frequency terahertz sensing applications, biomedical diagnostics, environmental monitoring, and security imaging.

ABSTRACT. In the ever-evolving landscape of technology, one paradigm-shifting innovation is emerging at the forefront of wireless communication and electromagnetic wave manipulation: Reconfigurable Intelligent Surfaces (RIS). This cutting-edge technology promises to revolutionize the way we interact with and harness electromagnetic waves, opening doors to unprecedented capabilities in wireless communication, energy transmission, and beyond. RIS is a technology that involves using specially designed surfaces to manipulate and control electromagnetic waves, such as radio waves, microwaves, or even light. These surfaces are typically composed of small elements or units that can be electronically controlled to alter the reflection, transmission, and absorption properties of incoming waves. As a transformative technology, RIS introduces a new era of dynamic control over how we interact with the electromagnetic spectrum. RIS practical deployment is hindered by significant challenges, including limited operational bandwidth, performance degradation at oblique incidence angles, and polarization sensitivity. This talk gives a brief introduction to RIS technology, their capabilities and theory behind their operation. It will address several critical gaps and introducing a systematic, physics-based design framework that integrates rigorous analytical modeling with efficient numerical optimization.

ABSTRACT. In the contemporary global economy, a nation's competitiveness is no longer measured solely by traditional factors such as natural resources, labor costs, or industrial output. Instead, digital readiness and technological infrastructure have emerged as decisive determinants of economic standing. Among these, communication networks constitute a fundamental pillar within Global Competitive Indicators (GCI), directly influencing metrics such as innovation capacity, business dynamism, infrastructure quality, and technological adoption. Without advanced, resilient, and ubiquitous communication networks, countries cannot achieve the high competitiveness rankings necessary to attract investment, foster entrepreneurship, or participate effectively in international markets. Communication networks underpin multiple dimensions of the GCI framework. First, they enable high-speed data exchange and digital services, which are essential for knowledge-intensive industries, e-commerce, and financial technologies. Second, network quality measured by coverage, latency, bandwidth, and reliability directly affects productivity across all sectors, from manufacturing to logistics to education. Third, the widespread availability of broadband and mobile connectivity reduces information asymmetries, allowing small and medium enterprises to compete globally. Consequently, nations with superior communication infrastructure consistently outperform others in GCI sub-indices related to infrastructure, technological readiness, and innovation. Furthermore, the strategic importance of communication networks within GCI is amplified by emerging technologies such as 5G, fiber optics, satellite internet, and software-defined networking. These technologies not only enhance existing economic activities but also enable entirely new industries, including autonomous systems, telemedicine, smart agriculture, and immersive digital environments. Economies that lead in network deployment gain significant competitive advantages, including faster time-to-market for digital services, lower transaction costs, and greater resilience against disruptions. However, challenges persist. The digital divide both within and between nations remains a major barrier to achieving equitable competitiveness. Cybersecurity vulnerabilities, regulatory fragmentation, and insufficient investment in next-generation networks further hinder progress. Addressing these requires coordinated policies, public-private partnerships, and international cooperation. In conclusion, communication networks are unequivocally a key pillar of Global Competitive Indicators. Their quality, reach, and intelligence directly determine a nation's ability to compete, innovate, and prosper in the 21st-century digital economy.

ABSTRACT. Autonomous robotics has revolutionary possibilities for visitor engagement and education when integrated into cultural heritage organizations. However, the implementation of smart museum guides is largely dependent on reliable computer vision systems that can recognize artifacts accurately and in real time. A deep learning framework for the multi-class categorization of Egyptian artifacts that is suited for intelligent museum robotics is presented in this paper. Three state-of-the-art architectures, ResNet18, EfficientNet-B0, and Vision Transformer (ViT-Base), were empirically evaluated in comparison using a carefully selected dataset of 4,782 photos from the Egypt Monuments Dataset. The platform enables automated, context-aware visitor assistance by enabling autonomous robots to identify important Egyptian artifact types, including statues, temples, pyramids, and royal busts. EfficientNet-B0 outperformed ViT-Base (96.3%) and ResNet18 (91.1%), according to experimental results, with a validation accuracy of 96.9%. Additionally, EfficientNet-B0 is the most practical model for real-time deployment in robotic systems with limited resources because it provides an ideal balance between accuracy, inference speed, and computational efficiency. This work advances intelligent human-robot interaction in museum settings and automated cultural heritage preservation.

ABSTRACT. Traffic loading, roadway aging, and inherent pavement defects significantly contribute to the development of cracks, which in turn adversely affect transportation infrastructure safety and performance. This forces civil engineers to spend significant amounts of time, effort, and cost on inspection, classification, and maintenance. Conventional crack detection relies on manual surveys or offline image analysis, which are labor intensive, time-consuming, and unsuitable for real time decision making. This research presents a low-cost autonomous robotic system that detects and classifies road surface cracks in real time according to standard engineering categories. Using embedded artificial intelligence and computer vision, the robot monitors pavement conditions and makes independent navigation decisions. It wirelessly transmits structured reports to a central platform that allows users to interact with all aspects of the project comprehensively and effectively. Field experiments demonstrate classification and autonomous navigation for monitoring. The autonomous road inspection system (ROVIL), which combines cloud-based monitoring, real-time wireless communication, and embedded deep learning into a single smart infrastructure platform, is presented in this study. A lightweight CNN model is used for crack detection and classification on device, with up to 98% confidence in real-world inference and 94.6% validation accuracy. A centralized server receives inspection data in JSON format over Wi-Fi, stores it in a SQL database, and displays the crack type, timestamp, confidence score, and GPS location on a web-based dashboard. This system is useful for smart transportation applications, the suggested system exhibits dependable, scalable, and real-time road condition monitoring.

ABSTRACT. Intelligent Transportation Systems (ITS) require reliable vehicle detection, yet model accuracy is frequently compromised by environmental variables. To reduce indeterminacy in complex urban landscapes, this study proposes an improved framework that combines the YOLOv11 architecture with neutrosophic image preprocessing. The suggested hybrid model (YOLOv11 architecture with neutrosophic images) performed better in challenging conditions, such as poor lighting, rain, fog, dust, nighttime, and sunny conditions, when tested on the Smart City Cars Detection dataset. The results reveal an improved Micro-F1 score of 0.4937 and an improvement in mAP@0.5 from 0.5892 to 0.6206. The "Motorbike" class precision increased from 66.67% to 80.00%, indicating that the model successfully identified small items. The Neutrosophic-YOLO11 framework demonstrated higher recall and improved resilience against noise and glare, albeit at a slight trade-off in micro-precision, offering a scalable method for practical smart city applications.

ABSTRACT. Polarimetric synthetic aperture radar (PolSAR) image classification remains challenging under limited labeled data due to the high dimensionality of polarimetric scattering information and the tendency of deep neural networks to overfit. This paper proposes a lightweight hybrid framework that decouples feature learning from classification to improve robustness under low-label conditions. A compact MobileNet-inspired convolutional neural network (CNN) is employed to extract spatial–polarimetric features using depthwise separable convolutions that preserve channel-wise scattering characteristics while maintaining low model complexity. The learned feature embeddings are then classified using a Random Forest (RF) ensemble. The proposed framework is evaluated on four benchmark PolSAR datasets, Flevoland and San Francisco Bay scenes at both L-band and C-band frequencies, using the strict 2% labeled training protocol. Experimental results demonstrate that the proposed MobileNet–RF framework consistently outperforms the Adaptive CNN (ACNN) baseline, achieving up to 99.91% overall accuracy (OA) while significantly reducing computational and memory requirements. In addition, an analysis of spatial window sizes reveals that moderate spatial contexts provide an effective balance between classification accuracy and computational cost, with a 15×15 window offering an optimal accuracy–memory trade-off for practical deployment on standard hardware.

ABSTRACT. Sustainable agriculture depends on effective water management, especially in arid areas like Egypt's Nile Delta. This paper describes a complete smart irrigation system that combines a high-performance hybrid machine learning model for predictive soil moisture analysis with an inexpensive Internet of Things (IoT) sensor network. The hardware utilizes a sensor node topology communicating via Bluetooth Low Energy (BLE). Our proposed solution is a hybrid system based on the Random Forest (RF) and the Long Short-Term Memory (LSTM) networks. The hybrid system, after being trained on 20 years of agroweather data, achieves a 16% improvement over the MAE of the LSTM approach. The hybrid system achieves a 23% reduction in water consumption.

ABSTRACT. Dorsal hand vein recognition is a reliable biometric approach for confirming identity; however, it is frequently hampered by issues such as lighting fluctuations, rotation, and occlusion. To solve these limitations, this research presents an improved Multibranch Lightweight Convolutional Neural Network (CNN) for vein recognition. The suggested design uses many parallel branches to capture different and distinguishable vein properties while drastically decreasing computational complexity and memory consumption. When tested on two publicly available datasets, the model obtained an impressive 99.9% accuracy, with 98% precision and recall, and a total processing time of 1086.33 seconds. Experimental results show that the proposed model surpasses state-of-the-art approaches in terms of accuracy, speed, and memory efficiency, demonstrating its effectiveness and applicability to real-world biometric applications.

ABSTRACT. Palm vein is widely adopted as a secure biometric modality due to the intrinsic difficulty of forging subsurface vascular patterns. While multi-spectral palm vein acquisition has been explored in previous studies, effectively combining spectral information with secure and efficient template protection is still a challenging problem. In this paper, a multi-spectral palm vein recognition system that combines complementary features from six spectral bands with cancelable biometric template generation is proposed. The MobileNetV2 lightweight network is employed for feature extraction from each spectral band. The model has a low computational complexity due to its depth-wise separable convolutions, making it well suited for real-time and resource-constrained environments. The extracted features are concatenated to enable effective exploitation of complementary spectral characteristics. The Relief-f feature selection algorithm is employed to model feature relevancy by selecting the most discriminative features according to their ability to distinguish between neighboring samples. Dimensionality is reduced and robustness to noise is enhanced without requiring complex optimization. Isotropic hashing (IsoHash) further transforms the features into compact binary codes with isotropic variance across hash dimensions. The algorithm supports fast Hamming-distance-based matching. It contributes also to template protection by producing non-invertible, revocable, and unlinkable templates. With compromised keys, the proposed system yields an identification accuracy of 98.5% and verification EER of 0.83% before hashing, versus 96.33% and 1.39% after hashing. A global measure D_↔^sys value of 0.05 is obtained, and thus correlation attacks are mitigated. These results exceed previous studies, offering higher accuracy alongside essential cancelable template properties.

ABSTRACT. Urban traffic congestion is a significant challenge in modern cities, affecting driver well-being, road safety, and transportation efficiency. Understanding the factors that contribute to driver stress is essential for developing effective traffic management strategies. This study proposes a machine learning–based approach to predict driver stress levels in urban traffic environments using the Smart City Traffic Stress Index dataset. The model utilizes multiple traffic-related features, including traffic density, signal waiting time, average speed, road quality, weather conditions, and driver experience level. To enhance the interpretability of the predictive model, Explainable Artificial Intelligence (XAI) techniques are applied. In particular, SHapley Additive exPlanations (SHAP) are used to analyze the contribution of each feature to the model’s predictions. Experimental results show that the proposed model achieves strong predictive performance with an R² score of 0.909. The explainability analysis reveals that traffic density, driver experience level, and signal waiting time are the most influential factors affecting driver stress. These findings provide valuable insights into how urban traffic conditions influence driver stress levels. The proposed approach supports the development of intelligent traffic management systems for safer and more efficient urban mobility.

ABSTRACT. A significant task in remote sensing is context-aware segmentation for high-resolution aerial images, which supports applications such as land cover classification, urban analysis, and environmental monitoring. In this paper, we propose a new framework directed to enhance the segmentation accuracy of aerial images by combining a split-depthwise convolution layer to handle high-resolution aerial images with lower computational complexity and a focus-based feature modulation module to emphasize spatial regions of interest. Our model was trained and tested on aerial images collected from geographically distinct areas within Alexandria. The results show that the model with our proposed framework exceeded the original U-net model in terms of accuracy and IoU value. Our framework provides an efficient solution for processing large-scale aerial image analysis and can be adapted for remote sensing applications, such as infrastructure development in Egypt, urban planning, and environmental monitoring.

ABSTRACT. This study investigates land use/land cover (LULC) dynamics in New Damietta City, Egypt, over the period 2018–2025 using Sentinel-2 MSI Level-2A imagery at 10 m spatial resolution and a GeoAI workflow implemented in ArcGIS Pro. Three classes were mapped, namely Built-up Area, Vegetation, and Bare Land. The methodology combined supervised classification, independent accuracy assessment, and post-classification change detection within a unified city-scale framework covering 4,395 ha. A comparative evaluation of candidate classifiers showed that Random Forest outperformed the tested alternatives, achieving an overall accuracy of 85.0% and a Kappa coefficient of 0.77, compared with 76.2% and 0.65 for SVM and 72.5% and 0.59 for KNN. Accordingly, Random Forest was retained as the final classifier for generating the 2018 and 2025 LULC maps. The results indicate that built-up area increased from 1,614 ha (36.7%) in 2018 to 1,698 ha (38.6%) in 2025, corresponding to a net gain of 84 ha, while bare land decreased by 111 ha and vegetation showed a net increase of 27 ha. Cross-tabulation further revealed that Bare Land to Urban conversion was the dominant transition pathway, accounting for 558 ha. These findings demonstrate that GeoAI-based classification in ArcGIS Pro provides a reliable and operationally practical framework for monitoring urban expansion and green-space dynamics in planned coastal cities.

ABSTRACT. Land surface temperature behavior in coastal cities is controlled by surface composition, emissivity, moisture availability, and maritime advection. New Damietta has undergone rapid conversion from coastal and sparsely vegetated surfaces to impervious urban fabric, altering radiative forcing and surface energy partitioning. This study analyzes seasonally matched summer Landsat 8 OLI/TIRS Collection 2 Level-1 scenes acquired in 2018 and 2025 to retrieve emissivity-corrected land surface temperature (LST) and to derive the normalized difference vegetation index (NDVI) and normalized difference built-up index (NDBI). The analysis addresses surface urban heat island behavior expressed by satellite-derived LST rather than near-surface air temperature. Results showed stronger 2025 thermal anomalies over zones of expanded built-up response and reduced vegetation. Pearson correlation and ordinary least squares regression confirm persistent positive LST-NDBI coupling and negative LST-NDVI coupling, with R2 values of 0.648 and 0.624 for LST-NDBI and 0.661 and 0.564 for LST-NDVI in 2018 and 2025, respectively. Because one summer scene was used for each epoch, the inter-annual mean LST difference is interpreted together with possible meteorological variability; the strongest evidence therefore lies in the repeated spatial coincidence of hotspots with higher NDBI and lower NDVI.

ABSTRACT. This paper is a comprehensive overview of the implementation of a remote sensing and perception stack for an autonomous low-speed vehicle (LSV) meant for complex environments within university campuses. In recent years, the application of the Global Navigation Satellite System (GNSS) for localization purposes has become the norm. This is due to the advantages offered by the system. However, the complex environments within university campuses pose significant challenges, especially due to signal degradation and multipath effects arising from the high-density infrastructure. In order to mitigate the effects arising from the radio-frequency limitations, the paper will analyze the hybrid sensing architecture, which includes ultrasonic sensors, camera-based semantic vision, and LiDAR-based 3D point clouds. This will allow for the effective perception of the environment. The paper will examine the trade-off associated with wave propagation and detection within different bands, particularly with reference to the night-time functionality of LiDAR, semantic vision, weather resistance, and radar. Moreover, the overview will delve into the application of sensor fusion techniques, incorporating both probabilistic Kalman filters and deep learning-based perception stacks, to ensure the overall robustness of the system. By incorporating the current advancements in Simultaneous Localization and Mapping (SLAM) techniques and Infrastructure-to-Vehicle (I2V) communication, the paper will establish the existing gaps within the application of remote sensing.