Task scheduling plays an important role in cloud computing, as it directly affects makespan, resource utilization, and energy consumption in data centers. With the increasing scale of cloud infrastructures, reducing energy usage and operational cost while maintaining efficient task execution has become a key research challenge. In this work, we propose a priority-aware task scheduling approach based on the Lion Optimization Algorithm (Priority-LOA), which includes both VM priorities and task priorities, which are modelled using the LOA. It is designed to minimize energy usage and power costs in data centers, ensuring efficient task-to-VM mapping. To achieve this, VM and task priorities are first computed and then we apply LOA to optimize energy consumption, makespan, power cost and resource utilization. The proposed scheduler is implemented in the SimPy simulation and evaluated using Google Cloud Jobs workloads ranging from 100 to 1000 tasks with task size 15,000 to 900,000. Experimental results demonstrate that the proposed Priority-LOA-based scheduler achieves near-optimal makespan by 42% and 56.63%, resource utilization by 54% and 109.70%, energy consumption by 31.74% and 41.78% and cost by 31.74% and 41.78% compared to Genetic Algorithm (GA) and Particle Swarm Optimization (PSO).

With the proliferation of Internet of Things (IoT) applications, a massive amount of data has been produced, requiring an efficient platform to store and process this data. Cloud computing has the ability to tackle such enormous data, but cannot provide real-time response to latency sensitive IoT applications. Fog computing delivers the cloud service at network edge with rapid response to IoT applications, but the computation offloading decision, unpredictable demands, unbalanced workload among fog servers and heterogeneity become challenging issues. Hence, this study applied an effective approach named Spider Monkey Optimization (SMO) to address the mentioned challenges and employs the proposed framework to ensure the Quality of Service (QoS) parameters performance. This framework uses an adaptive approach to allocate the workload based upon the computation ability of the resource, and continuously monitoring the workload among the fog nodes avoids the possibility of overloading and underloading the fog nodes. Exploration and exploitation ability of the SMO algorithm reduces the chances of being trapped in the local optimum and decides the optimal offloading destination. The performance of the proposed approach is assessed in a simulation environment, showing that the proposed algorithm reduces parameters such as latency, communication overhead, cost and energy consumption compared to baseline approaches Ant Colony Optimization (ACO), Particle Swarm Optimization (PSO) and Whale Optimization Algorithm (WOA) using the same experimental environments.

Traffic congestion in Indian cities causes annual economic losses exceeding Rs 1.5 lakh crore. This paper proposes the TRAC (Traffic Routing and Allocation Control) system combining YOLOv3 object detection, virtual lane adaptation, and PCU-weighted scheduling for unstructured heterogeneous traffic. TRAC achieves 43.2s average waiting time across 1600 SUMO simulation cycles, reducing waiting time by 43% versus traditional methods (74.25s) and 27% versus actuated controllers (59s). Throughput increases 28% (1087 vs 862 vehicles per 10 minutes). The edge-deployable pipeline runs on NVIDIA Jetson Nano with 85ms end-to-end latency using a custom 12,500-frame Indian traffic dataset (0.87 mAP). Virtual lanes handle non-lane discipline while 8-class detection (car, bike, bus, rickshaw, etc.) enables accurate PCU weighting. This represents the first system combining PCU-weighted optimization with virtual lane adaptation specifically designed for chaotic Indian traffic conditions.

Multi-criteria decision making (MCDM) often faces challenges in dealing with data with different scales and units, as well as ensuring the stability of ranking results against changes in criterion weights. This article proposes a new framework called multi-objective hybrid normalization for ideal solution-based alternative ranking (MONAS) that integrates hybrid normalization techniques and ideal solution approaches to improve the accuracy and consistency of alternative rankings. Weight uncertainty in MCDM reflects the variability or ambiguity in determining the relative importance among criteria, which can affect the stability and reliability of the final decision outcomes; however, this condition can be minimized through the MONAS approach, which adaptively integrates multi-objective normalization, thereby balancing the influence among criteria and improving the consistency of ranking results. The normalization hybrid used in this study combines three approaches: min-max normalization, vector normalization, and sum normalization. These are integrated by calculating the average of the results of these three normalization methods for each data value.  
The data domain used in the MONAS method for evaluating supplier performance includes seven alternatives and six assessment criteria. These criteria represent a combination of benefit and cost attributes, with value distributions that vary across them. Experimental results show that MONAS is capable of producing more reliable rankings that are resilient to data uncertainty compared to conventional methods. Testing results using 10 popular MCDM methods indicate that MONAS can provide more stable and reliable alternative rankings compared to those conventional methods. This framework offers effective and adaptive solutions for various complex decision-making applications across different domains, thereby improving the quality and trust in the decision-making process.

The new hierarchical path planning and tracking control model for autonomous vehicles utilizes hybrid optimization techniques, an adaptive potential field approach, and reinforcement learning to achieve safe, efficient, and adaptive path planning of the vehicles. Initially, a dynamic environment is generated to process path planning and traffic control with obstacles. The global path is computed using the Hybrid Dung Beetle and Hippopotamus Optimization (HDBHP) algorithm for minimum path distance, minimum obstacles, and minimum energy. This path is smoothed using Robust Locally Weighted Regression with Curvature Smoothing (RLWR-CS) to fulfill the vehicle kinematic constraints and obtain smoothness. For real-time local modulation, the Tent Map-Artificial Potential Field (TM-APF) approach is used, which is sensitive to obstacles and uses a chaotic path to plan for a better response. Control is done through Goal-Conditional Q-learning (GCQL) and Prioritized Q-learning (PQL), where decision-making is made with specific goals and prioritized experience replay. An integrated reward function ensures that the path is accurate, safe, fast, and comfortable for the passengers and ensures that the vehicle adapts to the roads from the start to the destination. Simulation results demonstrate that the proposed approach achieves improved Cumulative Rewards of 171.704 at episode 968 and 196.141 at episode 9,251. Also, the presented approach has a lower execution time of 1.65 seconds and outperforms existing approaches such as Double Deep Q-Network (DDQN), Proximal Policy Optimization (PPO), Deep Q-Networks (DQN), and Soft Actor-Critic (SAC).

A method for semantic graph analysis of the evolution of internet fraud is proposed based on the construction and comparative analysis of temporal semantic networks. The study is conducted using a corpus of 51,755 thematic documents retrieved from the InfoStream information system covering the period 2011–2025. A Retrieval-Augmented Generation (RAG)-based workflow was employed to retrieve relevant document contexts and support their semantic processing during network construction. Temporal semantic networks representing the conceptual structure of internet fraud at successive stages of its evolution were constructed for three five-year intervals, together with an integral semantic network characterizing the stable conceptual core of the subject domain.
The topological characteristics of the networks were analyzed using the framework of graph theory—specifically degree, betweenness, and eigenvector centrality—enabling the quantitative assessment of structural changes, the identification of concepts with a high transformational role, and the tracking of thematic cluster evolution. It is demonstrated that the development of internet fraud is predominantly evolutionary in nature, occurring through the restructuring of existing semantic structures rather than their complete replacement.
Building upon this foundation, a hypothesis of structural continuity for semantic networks is formulated, and a model for forecasting their evolution is proposed. This model involves estimating the structural transformation operator, predicting the emergence of new concepts and semantic relationships, and constructing the forecasted network for the subsequent time interval. The validity of the proposed approach is confirmed through retrospective validation, which entails forecasting semantic networks for the subsequent period using solely data from preceding temporal slices, followed by a comparison between the forecasted and actual structures. The obtained results demonstrate the potential of utilizing temporal semantic-graph analysis as a tool for the early detection of structural changes in the information space and the forecasting of new trends in the development of internet fraud. The retrospective validation confirmed the model's high predictive capability, achieving a Precision of 0.870, a Recall of 0.816, and an F1-score of 0.842 for the 2021–2025 forecast interval, alongside a strong rank correlation (ρ = 0.811) in preserving node centrality. These quantitative results demonstrate the practical validity of temporal semantic-graph analysis as a tool for the early detection of structural changes and the forecasting of new trends in the development of internet fraud.

Inaccurate classification of the malaria burden on households hinders precise evidence required for effective malaria control through strategic interventions, early detection of risk, optimal allocation of resources, informed decision-making, and formulation of appropriate policies; thus, raising the global malaria burden to approximately 95%. Addressing this menace requires the development of a tree-based ensemble model for precise classification of malaria burden on households, which utilizes characteristic features of data associated with malaria burden on households obtained from the repository of Malaria Indicator Survey of 2015 and 2021 and the Demographic and Health Survey of 2018 respectively, covering six geopolitical zones of the selected households in urban and rural Nigeria. The model incorporates majority voting, k-fold cross-validation and grid search algorithms for optimal model tuning and was implemented in the Python programming language using relevant features of data sourced from these repositories. The model performance was evaluated with precision, recall, F1-score, and Area under Curve (AUC). The overall mean accuracy achieved by the ensemble model classifier was 90.34%, indicating a high level of predictive performance and the model’s suitability for the precise classification task, and outperforming the other base models such as XGBoost (89.61%), Gradient Boosting (93.26%), CatBoost and LightGBM (89.60%), AdaBoost (87.69%). Nevertheless, the significant result is primarily driven by the high feature importance scores of the three most influential features, namely H20 (26.8%), H25 (24.5%), and H26 (21.2%). However, this model offers a practical tool for precise classification, enabling field personnel to identify high-priority areas for intervention. This tool can serve as a valuable asset for government agencies and policymakers by facilitating evidence-based decision-making and the formulation of targeted strategies to alleviate malaria burden on households. By harnessing data-driven insights, interventions can be more precisely directed toward the most vulnerable populations, thereby supporting the eradication of the disease on households. This study would further contribute to the achievement of Sustainable Development Goal (SDG) 3: ensuring healthy lives and promoting well-being for all.

The tourism industry has experienced substantial growth in recent years, propelled by the evolving customer preferences and pervasive influence of social media platforms. To gain a competitive advantage, it is imperative to develop innovative approaches that enable the delivery of personalized services and a deeper understanding of customer satisfaction. By leveraging user-generated content, firms can better understand dynamic customer trends and optimize their marketing strategies accordingly. However, our research aims to develop a novel customer segmentation framework using advanced deep learning techniques to analyze user-generated content within the rapidly growing tourism industry. The methodology employs a novel hybrid deep learning approach that integrates Sentence-BERT for contextualized embeddings and a GRU-based Autoencoder for feature reduction in an adaptive manner. This process is followed by k-means clustering, which segments customers using both multi-criteria ratings and online reviews to provide a holistic understanding of customer experiences. The proposed framework's effectiveness is compared with multiple state-of-the-art models using robust evaluation metrics such as the silhouette coefficient and the Davies–Bouldin index. The study’s findings highlight that the proposed model classifies customers into four distinct segments, and the statistical test performed shows the significance of our result. The research study contributes to the advancement of user-generated content based market segmentation in the tourism industry by unifying textual and numerical feedback into a single analytical framework. The findings offer valuable insights for tourism businesses seeking to enhance customer satisfaction through personalized service strategies and data-driven decision-making.

This study investigates computational thinking (CT) proficiency in programming-based learning environments using a unified analytical framework that combines statistical analysis and predictive modeling. CT proficiency is examined across five pedagogically grounded dimensions—abstraction, decomposition, algorithmic thinking, debugging, and pattern recognition—derived from rubric-based assessment of student work. First, descriptive and inferential statistical analyses are conducted to examine overall CT performance and gender-related patterns, providing correlational insight into group-level differences. Building on this analysis, a transformer-based model (BERT) is employed to predict continuous CT proficiency from students’ code comments and reflective journals, enabling semantic modeling of higher-order reasoning expressed in natural language. The predictive performance of BERT is benchmarked against traditional machine learning and lightweight deep learning baselines. Results show that transformer-based semantic modeling improves predictive accuracy while maintaining interpretability through post hoc explanation methods. Explainable AI techniques are used to identify linguistic and behavioral indicators associated with CT proficiency, and gender-related interpretations are derived through subsequent comparative analysis rather than direct prediction. Overall, the study positions deep learning as a complementary tool to statistical analysis for understanding and predicting CT proficiency.

Autism spectrum disorder (ASD), a complex neurodevelopmental disorder, is typified by social interaction challenges, communication difficulties, and repetitive activities. Effective intervention requires an early and precise diagnosis. This paper introduces NeuroASD-Net, a novel framework for ASD detection from structural Magnetic Resonance Imaging (MRI). The approach integrates advanced preprocessing techniques such as histogram matching and Contrast Limited Adaptive Histogram Equalization (CLAHE) to standardize input data. The next stage is the segmentation of key brain regions, like cerebrospinal fluid, white matter, and gray matter, using k-Means clustering. This segmentation step isolates critical regions for subsequent analysis. Following segmentation, a Two-Level Feature Extraction (TLFE) model is applied. In the first level, spatial and textural patterns are extracted using Local Binary Patterns (LBP), Gabor filters, and the Gray Level Co-occurrence Matrix (GLCM), capturing essential morphological characteristics of the brain. Feature selection is then optimized using the Self-Adaptive Black-Winged Kite Algorithm (SA-BKA). In the second level, high-level features are extracted using EffiDSR-Net, in which the EfficientNet model is enhanced with Dual Scale Residual (DSR) Blocks and a Convolutional Block Attention Module (CBAM). The features from both levels are fused to form a comprehensive feature set. Finally, classification is performed using a Softmax layer, achieving precise ASD detection. The proposed framework demonstrates enhanced diagnostic accuracy, achieving an Accuracy of 99.28%, Sensitivity of 99.31%, Specificity of 99.24%, F1-score of 99.31%, and an Area Under the ROC Curve (AUC) of 0.995, indicating its effectiveness for clinical ASD detection.

To address the “information gap” in social work arising from unstructured data and limited labeled instances, this study proposes a semi-supervised learning framework based on a Teacher-Student BERT architecture. Based on an analysis of 50,000 case records spanning 2019 to 2024, the model incorporates Latent Dirichlet Allocation for topic extraction alongside a multidimensional gap analysis. The proposed method attained a state-of-the-art F1 score of 0.913, markedly surpassing baseline BERT models. Notable findings include a 67.6% increase in mental health-related discourse and the quantification of significant systemic deficiencies, such as inadequate coverage in elderly care (45.8%) and substantial unmet needs in medical assistance (46.4%). Furthermore, to address the profound challenges of class imbalance and the data-hungry nature of transformer models, this study integrated generative artificial intelligence (AI) data augmentation techniques. This approach produced synthetically varied case narratives that preserved the original socio-economic context while expanding lexical and structural diversity, thereby significantly enhancing the classification accuracy of minority classes. Additionally, the use of a Mean Teacher denoising framework and knowledge distillation drastically reduced the computational inference time, rendering the architecture highly suitable for deployment in resource-constrained social welfare environments. Prior to analysis, all case records underwent rigorous automated Personally Identifiable Information (PII) scrubbing utilizing the Microsoft Presidio framework to guarantee data privacy. To ensure reproducibility, the Teacher-Student training codebase, prompt architectures, and anonymized synthetic data samples will be made publicly available upon request. This research effectively transforms administrative textual data into actionable strategic intelligence, offering a scalable and evidence-based tool to enhance resource allocation and inform policy development.

Automatic text summarization plays an important role in transforming lengthy documents into concise and informative representations. The abstractive summarization aims to generate sentences that capture the semantic meaning of the source text. This study proposed semantic and syntactic augmentation in transformer-based abstractive summarization using two encoder–decoder models such as a BART-based model enhanced with SpaCy Part-of-Speech analysis and RoBERTa-based semantic embeddings, and a prompt-guided T5 model, in which task-specific textual prompts were appended to the input sequence to guide the summarization process during fine-tuning and inference. Both models were trained and evaluated on the BBC News dataset using ROUGE parameters. The final results show that the T5 model obtained ROUGE-1 of 45.61, ROUGE-2 of 30.12, and ROUGE-L of 45.12. The BART model obtained ROUGE-1 of 44.28, ROUGE-2 of 28.70, and ROUGE-L of 44.72. The results indicate that the prompt-based T5 model combined with linguistic feature augmentation can improve the quality of abstractive summaries. However, as the reference summaries in news datasets contain extractive characteristics, ROUGE-based evaluation reflects lexical overlap in addition to abstractive generation. The final findings show the improved abstractive summarization performance within the constraints of the dataset and evaluation protocol.