Emotion estimation from face expression analysis has been extensively examined in computer science. In contrast, classifying expressions depends on appropriate facial features and their dynamics. Despite the promising accuracy results in handled and favorable conditions, processing faces acquired at a distance, entailing low-quality images, still needs an influential performance reduction. The primary objective of this study is to introduce a Real-Time Emotion Recognition system-based Fog Technique, which was developed to track and observe human emotional states in real time. This paper provides a comprehensive integration of PCA-based feature selection with a specific version of YOLO (YOLOv8), in addition to spatial attention for real-time recognition. The developed system demonstrates superiority in edge deployment capabilities compared to existing approaches. The proposed model is compared with the CNN_PCA hybrid model. First, Principal Component Analysis (PCA) is employed as a dimension-reduction tool, focusing on the most informative characteristics during training, and then CNN as classification layer. The proposed system's performance is assessed via a dataset of 35,888 facial photos classified into seven classes: anger, fear, happiness, neutral, sadness, surprise, and disgust. The constructed model surpasses established pre-trained models, such as VGG, ResNet, and MobileNet, with different evaluation metrics. First, the PCA_CNN model achieved superior accuracy, precision, recall, and Area Under the Curve (AUC) scores of 0.936, 0.971, 0.843, 0.871, and 0.943.YOLO v8 aith attention model achieved 0.986, 0.902, 0.941, and 0.952. Additionally, the model exhibits significantly faster processing time, completing computations in just 610 seconds than other pre-trained models. To validate the model's superiority, extensive testing on additional datasets consistently yields promising performance results, further validating the efficiency and effectiveness of our developed model in real-time emotion recognition for advancing affective computing applications.

This work presents the design, analysis, of a floral ultra‐wideband (UWB) monopole patch antenna for wireless communication applications. The antenna is fabricated on an FR4 substrate with a dielectric constant of 4.4 and a thickness of 1.6 mm. Its wedge‐shaped ground plane structure features three floral radiator elements, with a V‐slot in the center region of the middle radiator having an 8‐mm radius and the two side radiators each having a 5‐mm radius. Comprehensive simulation studies reveal that the designed antenna exhibits an operational bandwidth of 2.37–13.96 GHz, resulting in a percentage bandwidth of 142% for S 11 < −10 dB. Additionally, the antenna achieves a peak gain of 5.23 dB, indicating excellent impedance matching and power transfer efficiency. The introduction of this innovative floral monopole UWB antenna represents a novel contribution in UWB antenna design, making it a promising candidate for a wide range of high‐speed wireless communication applications that require exceptional bandwidth, gain, and impedance matching characteristics.

At modest nanoparticle concentration, Nanofluid show notable improvements in characteristics including the convective heat transfer coefficient, thermal conductivity that rises with nanoparticle volume percentage and diffusivity. Nanofluid offer significant potential for designing nanotechnology-based operational cooling systems for a variety of innovative applications. MHD radiative NF flow on a curvilinear permeable sheet under Darcy-Forchheimer porous resistance is examined in this work by using ANNs. Heat generation and thermal dissipation are included in the analysis, together with temperature-dependent thermal conductivity and variable mass diffusivity. Convective boundary conditions are incorporated. Brownian motion and thermophoretic effects are taken into account. This model also integrates Dufour and Soret effects, investigates binary chemical reactions and thermal radiation, and incorporates entropy generation by using ANNs. The PDEs are transformed into ODEs using similarity variables. LBP-ANNs are used to create and explain a framework for entropy generation analysis. ANNs are used to graphically analyze temperature, concentration, velocity, pressure, thermal dissipation and entropy rate. The temperature gradient, velocity gradient, pressure gradient, concentration gradient and thermal dissipation rate are all represented graphically. In recent work, LBP-ANNs is used to discussed the solution behavior of MHD radiative NF flow on a curvilinear permeable surface with entropy generation impact. To deduced the solution behavior off'(eta) , theta(eta), phi(eta), P(lambda), P'(lambda) and NG(eta), the number of parameters including Rd, M, lambda, Fr, Nb, Nt and Br varied. In each case, ANNs are used to display the graphical results of MSE, EH, TSF, FSF, RGN-A, solution evaluation, and AE findings. The P(eta) profile rapidly increases with growing differences of M& lambda. The P'(eta) profile declines as an increase in M& lambda. Thef'(eta) profile rises when there is an increase in Fr while it declines as an increase in M. The theta(eta) profile declines with increasing difference of Nb& Rd. The phi(eta) profile tends to decrease when the values of Nt increase while it increases with increasing difference of Nb. The entropy generation NG(eta) profile rises up with rising values of Br and Fr. The MSE consequences (testing, training, validation) for NF flow on a stretching curved sheet lies between 10-10 to 1000. The values of performance grids are lies between 10-10 to 10-09, while gradients values lie around 10-8 to 10-07 by using ANNs. The EHA range recorded around 10-07 to 10-05 for all twelve scenarios of for NF flow on a stretching curved sheet. The R-squared value is equal to 1 for all data sets of for NF flow on a stretching curved sheet. The AE for NF flow on a stretching curved sheet is noted between 1 x 10-04 to 10 x 10-05 for all twelve scenarios.

This paper presents a novel model termed as Optimized Quaternion Charlier Moments Convolutional Neural Network (QCMs-PSO-CNN) to develop an intelligent method of recognizing the visual speech accurately that is phrase or word that a person spoke in the video through Lip-reading. The proposed method put forward the Optimized QCMs using our enhanced particle swarm optimization (PSO). The suggested PSO is improvised on linear inertia weights and a sine–cosine learning factor, which effectively strengthens the optimization capability. This method suggested a custom-built shallow CNN which incorporates optimized QCM as a filter in the first layer in a novel way for better recognition ability of the method. The study shows that this method is a good solution for reducing the high video image dimension and gaining time for training. The proposed architecture QCMs-PSO-CNN found to classify the digits, letters, or words effectively. Three standard video datasets such as GRID, LRW, and GLips are chosen to evaluate the performance of the proposed method. The experimental analyses show that the proposed method attains excellent recognizing performance results compared to the other recently investigated approaches which made use of complex models and deeper architecture.

A compact four-element mmWave MIMO antenna designed using characteristics mode analysis (CMA) with wideband performance and high isolation is presented. The antenna is realized on a 0.254-mm Rogers RO5880 substrate with meandered conductive strips arranged in a window-shaped configuration and a partially truncated ground-plane notch. CMA is employed to observe the modal behavior of the structure, revealing that Mode 1 is the dominant radiating mode, while Modes 2 and 3 contribute to resonance formation and bandwidth enhancement. The single antenna element occupies a compact size of 12 & times; 14 mm2 (0.1305 lambda 0 & times; 0.1119 lambda 0, where lambda 0 is the wavelength at 28 GHz), whereas the four-port MIMO configuration measures 24 & times; 32 mm2 (0.224 lambda 0 & times; 0.229 lambda 0). A decoupling structure is introduced to suppress mutual coupling, achieving interelement isolation better than 20 dB. The proposed MIMO antenna operates over a wide impedance bandwidth from 24.5 to 33.5 GHz and exhibits a peak realized gain of 6.2 dBi. Due to its symmetric layout, the design is inherently scalable to higher order MIMO systems. Experimental results obtained from a fabricated prototype show good agreement with simulations, validating key MIMO performance metrics, including ECC, MEG, CCL, and DG, confirming the suitability of the proposed antenna for mmWave RF applications.

Financial markets exhibit complex, non-linear dynamics characterized by high volatility and uncertainty, making accurate stock movement prediction a challenging task. This study introduces FusionLSTM-CNF, a hybrid deep learning framework that integrates multi-modal data fusion, Long Short-Term Memory (LSTM) networks, and confidence calibration for stock movement prediction under uncertainty. Our model leverages a late fusion architecture, combining the outputs of three parallel LSTM sub-models trained on technical indicators, textual sentiment from financial news, and cross-asset correlation signals. A confidence-aware neural fusion (CNF) layer adaptively reweights modality contributions based on learned uncertainty estimates. We validate our model across multiple financial indices including S&P 500, NASDAQ, and FTSE 100. Experimental results show a 12.3% relative improvement in prediction accuracy over single-modal LSTM baselines and 23.7% reduction in prediction variance. Compared to recent state-of-the-art hybrid architectures, improvements are more modest (1.1–1.8% absolute accuracy gain) but statistically significant (Diebold-Mariano test, [Formula: see text]). The framework provides calibrated confidence estimates (Expected Calibration Error: 0.031) that enable confidence-filtered trading strategies with improved risk-adjusted returns (Sharpe ratio: 0.267 for top-40% confidence trades versus 0.131 unfiltered). This work contributes to financial artificial intelligence by demonstrating the practical benefits of unifying uncertainty quantification with heterogeneous time-series fusion, providing practitioners with both predictions and confidence estimates for risk-aware decision-making.

Cerebral palsy is a prevalent neurodevelopmental syndrome that disrupts motor development in children, making early detection vital for effective intervention. Traditional clinical assessments rely on subjective observations, often missing minor motor abnormalities until they become severe, typically after 12 months of age. This article presents a novel deep learning model, TransCP-Net (Transformer-based Cerebral Palsy Network), designed for early detection of infant cerebral palsy through spatiotemporal pose representation learning. The architecture employs hierarchical spatial and temporal attention to analyze complex motion patterns in video sequences, integrating multi-modal data for improved accuracy. TransCP-Net incorporates specialized preprocessing, including temporal smoothing and trajectory encoding, to enhance feature learning. Tests on 1370 infant movement videos yielded impressive results: 94.7% sensitivity, 92.3% specificity, and an AUC-ROC of 0.968, outperforming ten state-of-the-art methods. Notably, it achieved a sensitivity of 96.3% within the critical 9-15 weeks range of fidgety movements, enabling timely interventions. Attention visualization highlights key areas such as the hips and shoulders, reinforcing clinical relevance. TransCP-Net demonstrates effectiveness across diverse clinical settings, serving as a viable, non-invasive tool for early cerebral palsy detection.

This work presents a wideband millimeter-wave (mmWave) antenna designed and experimentally validated using the Characteristic Mode Theory (CMT). The proposed single-element is implemented on an ultra-thin 0.254 mm substrate and occupies a compact footprint of 8 × 6 mm2. The antenna achieves a wide impedance bandwidth from 22.5 to 39 GHz, corresponding to a fractional bandwidth of 53.66%, covering the major 28 GHz and 38 GHz 5G mmWave bands. CMT analysis is employed to investigate the dominant electric-dipole modes across various frequency bands. Further, by keeping the ground length constant at [Formula: see text], the monople structure is varied to suppress the magnetic-dipole modes, thereby improving the linear polarization behavior with electric-dipole modes. The CMT for the proposed single-element structure illustrates that modes [Formula: see text], [Formula: see text], [Formula: see text], and [Formula: see text] are contributing to resonance at 24, 28, 32, and 36 GHz, where the first modes in each band indicate the dominant mode with relatively higher magnitude. The summation of these multiple excited modes across multiple bands has led to a wideband resonance with a stable radiation pattern. To improve the radiation performance for high-gain mmWave applications, the optimized single element is extended into a 1 × 8 linear array configuration. The proposed array achieves a gain variation from 11 to 15.8 dBi across the operating band, with peak gains of 12.8 dBi at 28 GHz and 15.5 dBi at 38 GHz. The measured results obtained from the developed prototype show good agreement with the simulated responses, validating the proposed CMT-guided design approach. Due to its compact size, ultra-thin profile, wide bandwidth, high-gain array performance, and clear modal interpretation, the proposed antenna is a promising candidate for future 5G mmWave wireless communication systems.

The Karakoram Highway (KKH) is a critical high-altitude transportation corridor connecting and is frequently affected by landslides due to steep terrain, complex geology, intense rainfall, and increasing anthropogenic activities. Reliable landslide susceptibility mapping is therefore essential for hazard mitigation and infrastructure resilience along this corridor. In this study, landslide susceptibility was assessed along a section of the KKH using two ensemble Machine Learning (ML) models: Random Forest (RF) and Extreme Gradient Boosting (XGB). A landslide inventory comprising 447 events was classified into fall, flow, and slide types and combined with nonlandslide samples for supervised modeling. Fourteen conditioning factors derived from remote sensing, hydrological indices, and geological data were used. Multicollinearity was addressed using Variance Inflation Factor analysis. Models were trained and validated using a 70/30 train-test split, and performance was evaluated using accuracy, precision, recall, F1-score, and AUC-ROC. The models showed excellent predictive performance for slide-type landslides (AUC = 0.99 for both RF and XGB) and strong performance for fall-type landslides (AUC = 0.92 for RF and 0.89 for XGB). Flow-type landslides exhibited moderate predictability (AUC = 0.68 for RF and 0.65 for XGB). The combined (total) models achieved very high accuracy (0.95) and AUC values of 0.99 (RF) and 0.98 (XGB). XGB classified a larger proportion of the study area into very high susceptibility (281.99 km2) compared to RF (71.23 km2), while RF distributed more area into low-to-moderate susceptibility classes. Both models produced geomorphologically consistent susceptibility patterns aligned with known landslide-prone zones along the KKH. XGB emphasized localized high-risk hotspots, whereas RF provided a more conservative spatial distribution. The results highlight the strong influence of anthropogenic activities and terrain controls on landslide occurrence and demonstrate the suitability of ensemble ML methods for robust landslide susceptibility assessment in complex mountainous environments.

Paralyzed people must maintain proper posture to prevent respiratory problems, pressure sores, and muscular contractures. A paralyzed person may slip out of their wheelchair several times as a result of poor sitting position. To overcome these challenges, a novel Emperor Penguin optimized sensor-Infused wheelChair (EPIC) framework has been proposed to constantly monitor the real-time posture of the paralyzed person's health state. The proposed model aims to monitor wheelchair posture and health in real time, which can automatically detect posture issues, providing timely alerts and feedback to the user via a mobile application. The proposed framework utilizes the emperor penguin optimizer (EPO) algorithm for feature selection to improve the accuracy of posture detection. A deep maxout network (DMN) analyzes the features to predict the posture of the wheelchair user patient. The proposed EPIC framework has been assessed using a Python simulator. The effectiveness of the proposed EPIC framework has been determined using evaluation metrics, such as precision, specificity, accuracy, and sensitivity. The proposed EPIC technique advances the overall accuracy by 10.1%, 7.73%, and 2.84% better than posture recognition, graphic user interface (GUI), and independent component analysis-kantorovich distance (ICA-KD), respectively.

With the emergence of cloud computing, the Internet of Things, and other large-scale environments, recommender systems have been faced with several issues, mainly (i) the distribution of user-item data across multiple information networks, (ii) privacy restrictions and the partial profiling of users and items caused by this distribution, (iii) the heterogeneity of user- item knowledge in different information networks. Furthermore, most approaches perform recommendations based on a single source of information, and do not handle the partial representation of users' and items' information in a federated way. Such isolated and non- collaborative behavior, in multi-source and cross-network information settings, often results in inaccurate and low-quality recommendations. To address these issues, we exploit the strengths of network representation learning and federated learning to propose a service recommendation approach in smart service networks. While NRL is employed to learn rich representations of entities (e.g., users, services, IoT objects), federated learning helps collaboratively infer a unified profile of users and items, based on the concept of anchor user, which are bridge entities connecting multiple information networks. These unified profiles are, finally, fed into a federated recommendation algorithm to select the top-rated services. Using a scenario from the smart healthcare context, the proposed approach was developed and validated on a multiplex information network built from real-world electronic medical records (157 diseases, 491 symptoms, 273 174 patients, treatments and anchors data). Experimental results under varied federated settings demonstrated the utility of cross-client knowledge (i.e. anchor links) and the collaborative reconstruction of composite embeddings (i.e. user representations) for improving recommendation accuracy. In terms of RMSE@K and MAE@K, our approach achieved an improvement of 54.41% compared to traditional single-network recommendation, as long as the federation and communication scale increased. Moreover, the gap with four federated approaches has reached 19.83 %, highlighting our approach's ability to map local embeddings (i.e. user's partial representations) into a complete view.

This article presents the design of a high-gain, highly isolated four-element multiple-input multiple-output (MIMO) antenna system operating at the millimeter-wave (mm-wave) 28 GHz frequency band. The radiating element of the MIMO antenna comprises a 1 × 2 array of crescent-shaped patch elements backed by a full ground plane, having overall dimensions of 33 × 33 mm2 (3.08λ × 3.08λ), where λ is the wavelength at 28 GHz. A 0.787-mm-thick low-loss dielectric substrate having a relative permittivity (εr) of 2.2 and a loss tangent (tanδ) of 0.009 is used for the antenna design. It is observed that the designed radiating element provides resonance at 28.18 GHz and has a 6.21% fractional bandwidth (FBW) and a peak realized gain of 10.5 dBi. The radiation and total efficiency of the radiator are noted to be >85% and ≥75%, respectively, in the operating bandwidth. For polarization diversity and high isolation characteristics, the MIMO elements are arranged in an orthogonal manner, which helps achieve an isolation of >22.5 dB for adjacent elements and >32.5 dB for diagonally placed elements in the entire band of interest. Furthermore, MIMO system parameters such as envelope correlation coefficient (ECC) <0.003, diversity gain (DG) >10 dB, total active reflection coefficient (TARC) < −10 dB, and channel capacity loss (CCL) <0.5 bps/Hz are observed, which fall within the operational limits. Based on the achieved outcomes, the proposed design is well suited for mm-wave 28 GHz communication devices.

Smartphones have become indispensable in modern life, integrating advanced communication, computation, and multimedia capabilities into compact, user-friendly devices. This research presents eight-element MIMO antenna system at 3.5 GHz and 5.4 GHz with fractional bandwidths 9.1 % and 20.1 %. Isolation levels exceeding 14.5 dB between antenna elements ensure minimal coupling among radiating elements. The antenna's efficiency ranges from 50 % to 65 % in first and 60 % to 70 % in second resonance, ensuring reliable performance across its operating frequencies. The total size of the board is taken as 150 x 75 x 0.8 mm(3) while the size of the antenna is kept 18 x 4.5 mm(2) (0.21 lambda x 0.06 lambda). The prototype is fabricated and tested in an anechoic chamber, with far-field plots measured at 3.5 and 5.4 GHz, confirming the antenna's directive characteristics. Furthermore, hand-grip analysis was performed to validate the performance of proposed MIMO system in practical scenarios, showing results in good agreement with simulations. MIMO performance metrics were validated, and envelope correlation coefficient (ECC) values was noted below 0.05, ensuring excellent diversity performance. Other critical metrics were also analyzed, including mean effective gain (MEG) with values of-3.5 and-3 and channel capacity loss (CCL) < 0.4 for entire band of interest indicating minimal degradation of the communication channel due to the antenna design. These results highlight the proposed antenna's suitability for integration into modern smartphones, providing efficient dual-band operation, high isolation, and excellent MIMO performance. Its compact size and robust design make it an ideal candidate for next-generation wireless communication systems, ensuring reliable connectivity and improved data rates in 5G applications.

Alzheimer's disease (AD) is a progressive neurodegenerative disorder that leads to memory loss and a decline in cognitive abilities. It primarily affects older adults and is the most common cause of dementia. Using deep learning, models can analyze brain imaging scans to detect specific patterns and biomarkers associated with the disease. Supervised learning models achieve high accuracy rates, but they require a large amount of data sets and labelled medical images. Self-supervised learning can achieve high accuracy rates with fewer training data. This study proposes a self-supervised attentive feature learning network (SSA-Net) for classifying Alzheimer's disease. The proposed approach leverages self-supervised learning and attention mechanisms to enhance the accuracy and reliability of the classifying model. We employ ResNet-50, incorporating attentive activation, which replaces the ReLU activation, improving the ability of the neural model to focus on the most relevant features in the input medical images. We use SimCLR (Simple Framework for Contrastive Learning of Visual Representations) with the ResNet-50 backbone as a self-supervised learning framework that effectively learns high-quality visual representations in brain MRI (Magnetic Resonance Imaging) scans without labelling. We used the Kaggle Alzheimer's classification dataset (KACD) containing brain MRI scans for training and testing. Experimental results on the KACD dataset show that the proposed attentive self-supervised ResNet50 reached 99.7% classification accuracy compared to the traditional ResNet50 with 98.1% accuracy. Evaluation metrics show the effectiveness of the proposed SSA-Net for the efficient classification of Alzheimer's disease.

Real-world applications benefit greatly from aerial imagery. AVarious modern applications utilizing aerial images; however, these images are often low-contrast due to imperfect atmospheric conditions and limitations in the imaging systems. Many methods exist to enhance the quality of aerial images, yet not all of them capable of producing desired results. Some may have high complexity, and others may require numerous inputs. On the other hand, it is observed that a low-contrast impact that is difficult to prevent throughout the data collection process degrades the quality of aerial images a lot. AsAa result, in this paper a novel method for improving aerial image contrast has been presented. Hence a two-phase approach for increasing contrast and remapping the intensities of an aerial image to its native dynamic range has been presented in this paper. Additionally, a regularization technique is provided using the two-step regularization and mapping procedures. For image quality assessment (IQA), two performance assessment metrics; measure of enhancement (EME) and Structural SIMilarity (SSIM) have been suggested to measure the quality of the results of the proposed algorithm. The experimental results indicate that the proposed approach increases the contrast of aerial image substantially as compared with other widely contrast enhancement methods.

Document Clustering has attracted the interest of many researchers who have created several solutions to this problem by combining different techniques, models, and algorithms. While famous for its simplicity, the most commonly used algorithm, K-Means, suffers from issues such as finding the optimal value for k and random initialization of the centroids. In this paper, we propose a hybrid methodology combining K-Means++ with the metaheuristic algorithm PSO to overcome the challenges of both these algorithms. K-Means++ is a smart initialization technique that selects clusters based on probability function so that they are further away from each other. Two methods, Elbow Analysis, and Silhouette Score, have been used to find the optimal value of k. The research also attempts to assess the results of the best-performing feature extraction techniques and create a combined approach using their average. Although the combined approach did not outperform the existing methods, it paved the way for further advancement and exploration of merging feature extraction techniques. The experiments were performed on four datasets: 20 Newsgroups, Reuters, WebKB, and Wine, and the results were evaluated using three evaluation metrics: Purity, Silhouette Score, and Jaccard Index. The proposed approach achieved a Purity of 0.95, 0.91. 0.86, 0.89, the Silhouette Score was measured to be 0.5, 0.05, 0.05, 0.59, and the Jaccard Index was reported to be 0.04, 0.04, 0.05, 0.14 on Reuters, WebKB, 20 Newsgroups, and Wine datasets, respectively.

As drones continue to see widespread adoption across commercial, private, healthcare, and education sectors, their commercial use is experiencing rapid growth. To enhance drone-based package delivery efficiency and improve customer experience, the vast amount of flight and recharging data collected from drones and stations offers valuable opportunities for predicting both resource availability within the sky network and drones' capacity to complete delivery missions. However, the variations in regional regulations and privacy restrictions enforced by drone service providers lead to data heterogeneity, making centralized processing of flight and charging history impractical. Running machine learning models locally at the service provider level (drones and stations) addresses privacy concerns, yet processing the large volume and diversity of raw data remains a significant challenge. To deal with these issues, a collaborative learning approach based on historical delivery data presents an elegant solution. Aiming to offer predictive scheduling of drone delivery missions, while taking into consideration the complexity, heterogeneity, and dynamic nature of their flight environment, we propose a predictive and federated approach for the resilient selection and composition of drone delivery services, leveraging the strengths of federated learning (FL) to handle data privacy and heterogeneity. Our method utilizes a federated Recurrent Neural Network (FL-RNN) model that combines predictive capabilities with federated behavior, enabling collaborative forecasting and efficient mission scheduling in low-congestion regions based on the most reliable drone services. Additionally, an enhanced A* search algorithm is defined to identify the optimal delivery path by factoring in station overload probabilities. Besides computational efficiency, experimental results demonstrate the effectiveness of our approach, achieving a 16.66% improvement in prediction accuracy and an 8.89% reduction in delivery costs compared to non-federated and non-predictive solutions.

Vehicle-as-a-Service (VaaS) refers to on-demand rides and the sharing/offering of various kinds of intelligent transportation facilities (e.g., smart buses, electric vehicles, autonomous cars) to move from a source to a destination across one or several regions. Coupled with smart transportation systems–which are critical for addressing road network issues such as traffic congestion, parking shortages, and safety concerns–VaaS is increasingly being adopted in smart cities. For example, a user may specify their needs in terms of source and destination stations, time and cost constraints, as well as preferred transportation modes and services (e.g., only connected buses and cars). Such a user profile is evaluated against available VaaS options under current and anticipated urban network conditions. However, current solutions do not support the customization of VaaS compositions and often treat user requests as traditional vehicle routing problems. In a smart mobility context, however, processing VaaS requests involves not only finding the optimal transportation path that meets user constraints (e.g., time, cost, transfer stations) but also selecting the top-rated combination of available VaaSs (e.g., a sequence of smart buses) with respect to the user profile (e.g., connectivity needs, specific facilities) and the quality of smart urban services. To address these challenges, the goal of this paper is to develop a multi-modal recommender system that enables the personalized selection, composition, and scheduling of VaaS services while optimizing trip constraints (e.g., minimizing trip time and cost, and maximizing VaaS availability and reputation). As a multi-population technique, Composite Particle Swarm Optimization (CPSO) is applied to aggregate the optimal set of high-quality and high-coverage VaaSs with respect to the requested trip. The regions composing the trip are explored using a modified A* search algorithm to find the optimal local (partial) path in each traversed region of the smart urban network. Comparative experiments involving two metaheuristics, a greedy algorithm, and a fuzzy clustering technique demonstrate the efficiency and superiority of our CPSO-based approach, achieving approximately a 28

Distributed Denial of Service (DDoS) attacks pose significant threats to network security, disrupting critical services by overwhelming targeted systems with malicious traffic. In this study, a machine learning-based approach is proposed to classify DDoS attacks using multiple classification models, including Random Forest (RF), Naïve Bayes (NB), K-Nearest Neighbors (KNN), Linear Discriminant Analysis (LDA), and Support Vector Machine (SVM). The DDoS-SDN dataset was used for training and evaluation, with feature selection via Backward Elimination (BE) and hyperparameter tuning using Grid Search with 5-fold Cross-Validation (CV = 5). Experimental results demonstrate a significant improvement in classification performance after feature selection and parameter optimization, with RF achieving the highest accuracy of 99.99%. In this study, we propose a machine learning-based classification framework enhanced by feature selection and hyperparameter optimization techniques through employing Recursive Feature Elimination (RFE) and Grid Search .Our model based on Random Forest (RF) achieved a remarkable accuracy of 99.99%, outperforming other baseline classifiers, including Naive Bayes (98.85%), K-Nearest Neighbors (97.90%), Linear Discriminant Analysis (97.10%), and Support Vector Machine (95.70%). In addition to accuracy, the RF model also demonstrated superior F1 score, recall, and precision, each reaching 99.99%. These results validate the effectiveness of our optimization strategy in improving classification performance. The study highlights the effectiveness of feature engineering and model optimization in enhancing DDoS detection accuracy, making machine learning a viable solution for real-time cybersecurity applications.

Voice User Interfaces (VUI) in consumer electronics often operate in noisy environments where speech quality degrades significantly. We propose AV-Net, a lightweight audiovisual speech enhancement model that addresses this challenge through novel cross-attentional feature fusion. Our approach dynamically integrates audio and visual modalities using a computationally efficient architecture combining a convolutional encoder-decoder (5.2M parameters) with a P3D-ResNet18 video encoder. The key innovation is a cross-attention mechanism that learns inter-modal correlations while preserving modality-specific features, outperforming conventional fusion methods. Evaluated on TCD-TIMIT and AVSE3 datasets under challenging conditions (SNR <=-5dB), AV-Net achieves a PESQ of 2.56 (vs. 1.26 baseline) and STOI improvement of 22%, while maintaining realtime performance (RTF=0.11). The model demonstrates strong generalization to unseen speakers and diverse noise types, making it particularly suitable for resource-constrained edge devices in healthcare, automotive, and smart home applications where robust speech interaction is critical.