Social media information is proliferating on the internet every day in various contexts. However, there is difficulty in using audio, especially in Arabic, due to a lack of resources and the nature of regional accents and their complexities. All these parameters are not considered when using text. This paper presents three models created for sarcasm detection using text, audio, or a hybrid of both in Arabic, as well as deep learning and other audio-processing techniques. The results showed that using audio data mixed with text data to identify sarcasm in Arabic has promising results, with an accuracy of 0.93, a precision of 0.89, a recall of 0.92, and an F1 score of 0.91.

Accurate multivariate time series forecasting remains a fundamental challenge across critical domains including traffic management, environmental monitoring, and agricultural planning. While recent advances in deep learning have shown promise, existing approaches face inherent limitations: convolutional models struggle with long-range dependencies and cross-variable interactions, whereas pure attention-based architectures often overlook crucial local temporal patterns and suffer from quadratic computational complexity. To address these complementary weaknesses, we propose a novel hybrid architecture that systematically integrates Temporal Convolutional Networks (TCNs) for efficient local feature extraction with Transformer multi-head attention mechanisms for global dependency modeling. Our TCN-Transformer model employs dilated causal convolutions to capture hierarchical temporal patterns across multiple scales, followed by multi-head attention layers that learn cross-variable dependencies and long-range temporal relationships. We validate our approach across three diverse real-world domains: traffic volume prediction, air quality forecasting, and wheat productivity estimation across five Egyptian governorates. Experimental results demonstrate substantial improvements over established baselines, with the TCN-Transformer achieving coefficient of determination ( R^2 ) values ranging from 0.91 to 0.96 across all datasets, representing improvements of 5-13 R^2 = 0.96 . The model outperforms seven baseline approaches including statistical methods (VAR, SVR), recurrent networks (LSTM, GRU, BiLSTM), and individual deep learning components (TCN, Transformer), demonstrating robust generalization across temporal resolutions (7-day and 30-day windows), data characteristics, and application domains.

This study leverages multimodal Large Vision-Language Models (LVLMs), such as LLaVA, IDEFICS, and Flamingo, to enhance liver disease diagnosis in high-burden regions like Egypt, which reports the world's highest liver disease mortality rate and second-highest liver cancer mortality (WHO, 2022). By analyzing the Eurorad's medical imaging dataset and extracting insights from clinical notes, LVLMs achieve over 90 % accuracy, precision, and recall in liver tumor classification. Additionally, LVLMs translate complex medical terminology into plain language, improving patient comprehension. Experiments demonstrate LVLMs' efficacy in radiology applications, positioning them as viable decision-support tools for hepatology in resource-constrained settings. Index Terms-Large Vision-Language Models (LVLMs), Large Language Model (LLM), Large Language and Vision Assistant for BioMedicine (LLaVA-Med), Contrastive Language-Image Pretraining (CLIP), Eurorad Dataset

This paper presents two effective approaches for Extractive Question Answering (QA) on the Quran. It addresses challenges related to complex language, unique terminology, and deep meaning in the text. The second uses few-shot prompting with instruction-tuned large language models such as Gemini and DeepSeek. A specialized Arabic prompt framework is developed for span extraction. A strong post-processing system integrates subword alignment, overlap suppression, and semantic filtering. This improves precision and reduces hallucinations. Evaluations show that large language models with Arabic instructions outperform traditional fine-tuned models. The best configuration achieves a pAP10 score of 0.637. The results confirm that prompt-based instruction tuning is effective for low-resource, semantically rich QA tasks.

Question answering systems face critical limitations in languages with limited resources and scarce data, making the development of robust models especially challenging. The Quranic QA system holds significant importance as it facilitates a deeper understanding of the Quran, a Holy text for over a billion people worldwide. However, these systems face unique challenges, including the linguistic disparity between questions written in Modern Standard Arabic and answers found in Quranic verses written in Classical Arabic, and the small size of existing datasets, which further restricts model performance. To address these challenges, we adopt a cross-language approach by (1) Dataset Augmentation: expanding and enriching the dataset through machine translation to convert Arabic questions into English, paraphrasing questions to create linguistic diversity, and retrieving answers from an English translation of the Quran to align with multilingual training requirements; and (2) Language Model Fine-Tuning: utilizing pre-trained models such as BERT-Medium, RoBERTa-Base, DeBERTa-v3-Base, ELECTRA-Large, Flan-T5, Bloom, and Falcon to address the specific requirements of Quranic QA. Experimental results demonstrate that this cross-language approach significantly improves model performance, with RoBERTa-Base achieving the highest MAP@10 (0.34) and MRR (0.52), while DeBERTa-v3-Base excels in Recall@10 (0.50) and Precision@10 (0.24). These findings underscore the effectiveness of cross-language strategies in overcoming linguistic barriers and advancing Quranic QA systems

Large Language Models (LLMs) have transformed numerous fields by offering innovative solutions that drive advancements across a wide range of applications. However, their widespread adoption presents several challenges, including variations in architectures, limitations in processing capabilities, and high computational resource demands for training. Addressing these challenges is crucial for maximizing the benefits of LLMs while ensuring their responsible and efficient use. This paper reviews LLMs, focusing on their key characteristics and the factors that influence their performance. It examines several prominent families of LLMs and provides a comparative analysis of their properties. In addition, it explores the classification of LLMs based on criteria such as availability, context window, and model size. In addition, the study explores advanced fine-tuning techniques, including Parameter-Efficient fine-tuning (PEFT) and Low-Rank Adaptation (LoRA), that enhance the performance and efficiency of models. Furthermore, it reviews the wide-ranging applications of LLMs and evaluates the methodologies used to evaluate their effectiveness.

Ancient Egyptian hieroglyphs represent a writing system of profound historical importance, utilized for over three millennia to record religion, history, administration, and art on papyrus and stone. The task of accurately translating hieroglyphs from images remains a significant challenge, particularly with diverse layouts and orientations encountered on ancient artifacts. While prior research employed computer vision techniques for hieroglyphic recognition and translation, many approaches focused on fixed orientations and writing styles. This paper addresses these limitations by introducing a novel methodology for extracting and arranging hieroglyphic signs from real-world artifacts, irrespective of format or orientation. The proposed solution offers adaptability through two key innovations. First, Layout Analysis Stage processes complex text structures in various row and column combinations. Additionally, this study created the first Hieroglyphic Layout Analysis Dataset covering all main styles and artifact types. Second, a dynamic Sorting Function adjusts to multiple text orientations (left-to-right, right-to-left, top-to-bottom) without predetermined directionality. The "Layout Analysis, Segmentation, and Classification (LASC)" approach accurately extracts and orders hieroglyphic signs from diverse artifacts. LASC was evaluated against authentic ancient Egyptian artifacts, yielding promising results in transcribing hieroglyphic texts with complex layouts and orientations. To the best of authors' knowledge, no existing methodology exhibits this adaptability in handling hieroglyphic layout and orientation complexities.

Integrating artificial intelligence (AI) with healthcare revolutionizes diagnosis and patient care by improving clinical documentation, managing electronic health records, and interpreting medical images, ultimately improving accuracy and efficiency. In regions like Egypt, which faced the world’s highest mortality rate from liver disease and the second-highest deaths from liver cancer in 2022, advanced AI tools such as the multimodal Large Language and Vision Assistant (LLaVA) have significant potential to address global challenges in liver disease diagnosis. LLaVA has emerged as a critical tool to support the challenges in disease diagnosis. LLaVA can classify and describe patient scan images, extract information from clinical notes, and provide accurate diagnoses through effective prompt engineering. Additionally, it facilitates answering questions, summarizing data, and translating complex medical terminology into simple language, enabling patients to better understand their medical reports and diagnoses. This paper emphasizes LLaVA’s role in medical applications, focusing on prompt engineering to enhance the effectiveness of pre-trained models. Experiments using LLaVA demonstrated promising results for liver tumor classification with quantitative metrics such as accuracy, precision, and recall that exceeded 90%. These findings align with extensive research on medical image and text analysis, highlighting the potential of LLaVA architecture to assist healthcare professionals in diagnosing complex cases.

Hallucination detection in text generation remains an ongoing struggle for natural language processing (NLP) systems, frequently resulting in unreliable outputs in applications such as machine translation and definition modeling. Existing methods struggle with data scarcity and the limitations of unlabeled datasets, as highlighted by the SHROOM shared task at SemEval-2024. In this work, we propose a novel framework to address these challenges, introducing DeepSeek Few-shot optimization to enhance weak label generation through iterative prompt engineering. We achieved high-quality annotations that considerably enhanced the performance of downstream models by restructuring data to align with instruct generative models. We further fine-tuned the Mistral-7B-Instruct-v0.3 model on these optimized annotations, enabling it to accurately detect hallucinations in resource-limited settings. Combining this fine-tuned model with ensemble learning strategies, our approach achieved 85.5 accuracy on the test set, setting a new benchmark for the SHROOM task. This study demonstrates the effectiveness of data restructuring, few-shot optimization, and fine-tuning in building scalable and robust hallucination detection frameworks for resource-constrained NLP systems.

Ammonia production is vital to the global chemical industry. Its CO2 removal unit remains an energy-intensive bottleneck, improving sustainability and operational costs. Traditional control strategies struggle with nonlinear dynamics and multivariable interactions, leading to inefficiencies i n CO2 removal processes. This study introduces a machine learning (ML) enabled framework tailored for CO2 removal in ammonia systems using Hot Potassium Carbonate (HPC) solution. Advanced ML models such as Linear Regression, Artificial Neural Networks (ANN),and XGBoost are employed to predict CO2 removal efficiency and energy consumption. E valuated on real operational data from an Egyptian ammonia plant, ANN and XGBoost achieved R2 scores of 0.99 for energy savings and over 0.82 for energy consumption forecasts. The framework reduced steam usage from 12.4 ton/h to 10.1 ton/h, achieving an 18.5% energy saving, and improved CO2 monitoring efficacy from 88.2% to 94.7%, a 6.5% increase. These results demonstrate the potential of ML to enhance energy sustainability and CO2 removal efficiency, addressing the unique challenges of HPC solution based systems and advancing sustainable industrial practices.

Fuzzy Formal Concept Analysis (FFCA) is a robust mathematical tool for analyzing data, particularly where uncertainty or fuzziness is inherent. FFCA is utilized across various domains, including data mining, information retrieval, and knowledge representation. However, fuzzy concepts extraction is a crucial yet computationally intensive task. This paper addresses the challenge of time efficiency in extracting single-sided fuzzy concepts from large datasets. A parallel algorithm is proposed to reduce computational time and optimize resource utilization, thus enabling the scalable analysis of expanding datasets. By computing fuzzy concepts across multiple threads in parallel, each thread processes an attribute independently to extract fuzzy concepts, which are then merged in the final step. The proposed algorithm extracts fuzzy-crisp concepts, which are more concise than other types of fuzzy concepts. Experiments were conducted to evaluate the performance of the proposed parallel algorithm against existing sequential methods. Experimental results demonstrate significant gains in computational efficiency, with the algorithm achieving an average time reduction of 68% compared to the attribute-based algorithm and up to 83%-time reduction compared to the fuzzy CbO algorithm across various types of datasets, including binary, quantitative, and fuzzy.

The integration of artificial intelligence (AI) in healthcare is revolutionizing diagnosis and patient care by improving clinical documentation and the management of electronic health records that depend on medical image interpretation, increasing accuracy, and reducing time. Egypt ranks first in liver disease and second in liver cancer mortality worldwide in 2020. Large language models, a subset of AI techniques, can assist in disease diagnosis. LLM models with multimodal capabilities can classify and describe patient scan images and extract information from clinical notes. These models can extract vital diagnoses with the support of prompt engineering, as one of these models can answer questions, summarize information, and translate complex medical terminology into plain language, enabling patients to understand their medical reports and diagnoses. There are two primary approaches to achieving this. First, fine-tuning can adapt the model to medical data, which can be resource-intensive. The second approach, pre-trained LLM models can be utilized to leverage pre-trained models to perform the necessary tasks, focusing on effectively using prompts to guide the model for precise and relevant outputs. This study highlights the role of generative AI models by focusing on prompt engineering, and how carefully crafting prompts can enhance the effectiveness of LLM models in medical applications with high accuracy. It demonstrates this through experiments using pre-trained models based on semantic similarity with GPT-4o and BioGPT. Implementing a zero-shot model for liver tumor classification is one of the prompt engineering techniques. The performance metrics achieved were impressive, accuracy, precision, recall, and F1-scores are 88, 81, 88, and 83 percent, respectively.

The recent mainstreaming of Artificial Intelligence (AI) applications and tools has significantly enhanced EFL educators’ research writing process. Nevertheless, few studies exist regarding how EFL educators understand AI applications and their integration into research writing processes and techniques. The current study aims to investigate the attitudes and perceptions of (n= 40) EFL teachers at Majmaah University concerning integrating AI tools in the Research writing process, by collecting data from mixed-methods source questionnaires and interviews. Thematic analysis was employed for qualitative data, while descriptive and inferential statistics through SPSS were used for quantitative data analysis. The questionnaire results demonstrated that most respondents have positive perceptions toward integrating AI tools, believing they can improve efficiency and quality in the research writing process. However, more familiarity with existing AI applications is needed to ensure this integration process. Moreover, the study affirms the significant impact of training and support on the effective integration of AI tools. Respondents strongly agree about the benefits of AI applications in streamlining literature reviews, aiding in data analysis, reducing errors, and enhancing overall language quality. Interview responses further emphasize the possible benefits of AI tools in the research writing process, highlighting efficiency gains, assistance in various writing tasks, and improved quality of EFL teachers’ research articles. All the participants stress the importance of ethical utilization, maintaining quality, and vigilance when integrating these tools into the research writing process.

Extracting unique and distinctive traits is one of the most important challenges that researchers face, who rely on biometrics to extract exceptional traits for an individual. A large amount of biometric evidence that can be identified and found in various research has been done. In this paper, a biometrics system is proposed that combines the benefits of fingerprinting and uses a novel strategy to combine it with the image-based fingerprint vein feature set. The proposed system is fast and performs effective personal identification by combining both features. The features extracted from the venous print and fingerprint are matched to the nearest neighbors of the authorized person forms to verify the identity of the person. Several experiments have been performed on selected datasets to evaluate the performance of the new biometrics system. The obtained results prove that our proposed system is superior to biometric systems that use the feature of single biometrics. However, our goal is to set up an algorithm that is inexpensive in terms of time complexity while keeping it at the required security levels.

Brain tumors are one of the most serious kinds of cancer that arise from the unchecked growth of brain cells inside the skull. Early and effective diagnosis of brain tumors is critical as it is a life-threatening illness. In recent decades, Deep Learning methodologies have gained significant traction for developing Computer-aided systems capable of swiftly and accurately diagnosing brain tumors. However, the primary challenge in leveraging robust deep learning models lies in identifying the most suitable classifier for a given classification task. This task is inherently demanding and time-consuming, involving the identification of optimal network architectures and carefully fine-tuning numerous hyper-parameters. To address these challenges, this paper proposes several ensemble methods aimed at fusing multiple baseline deep learning models in the domain of brain tumors. To accomplish this, baseline convolution-based models are trained, including the proposed CNN model. Various ensemble techniques, such as meta-learning, voting, and weighted voting, are applied. An ensemble method is also introduced based on the fusion of extracted baseline features. Experimental results and comparative analysis on a brain tumor dataset demonstrate that ensemble methods enhance the performance of the baseline model, yielding an accuracy score improvement of 16% and 1.3%.

Hepatocellular carcinoma (HCC) stands as the pre-vailing primary liver tumor, marked by elevated mortality rates. Around 80 % of instances emerge within cirrhotic livers, presenting a considerable obstacle in determining suitable therapeutic approaches. HCC can be classified into various categories, and by utilizing Barcelona Clinic Liver Cancer (BCLC) and Child- Pugh score stages, survival rate and duration for patients can be determined, as well as appropriate treatment plans devised. Several contributions are made by this paper. Firstly, the study partners with a hepatologist to pinpoint pertinent patient biomarkers used in forecasting Child-Pugh scores and BCLC stages. Subsequently, leveraging these predictions of stages and scores, personalized treatment strategies are suggested for patients. The dataset from the National Liver Institute in Egypt, comprising 1108 of patient records with 22 features, is evaluated using various machine learning models such as Random Forest, Gradient Boosting, Decision Trees, and Support Vector Machines (SVM). Results show that the Random Forest model achieves 92.77% accuracy in predicting Child-Pugh scores, while Gradient Boosting achieves 83.82 % accuracy in predicting BCLC stages. Subsequently, treatment plans are determined based on these stage and score predictions.

Brain tumors are the leading cause of death world-wide. Deep learning has been successful in previous tasks like classification. However, it's being limited by the reliance on a single imaging modality which isn't enough, where a single modality can provide higher performance but is unreliable for accurate treatment and diagnosis. This study aims to improve brain tumor classification using deep learning and fusion techniques of multiple modalities. The study employs three fusion approaches: image-level fusion, feature-level fusion, and wavelet-based fusion. Extensive experiments were conducted on the BRATS2020 dataset. Initially, we train and evaluate the performance of 21 baseline models, encompassing 20 CNN-based architectures alongside the vision transformer model. Moreover, we identify the highest-performing models within each class for fusion. Furthermore, inspired by the baseline models, we dive deeper, introducing each modality as input to its respective best-performing model and fusing the outputs for multi-modality model-level fusion. Finally, we employ wavelet-based fusion to optimize information integration, implementing Discrete Wavelet Transform on our dataset. Model-level fusion outperformed image fusion across all evaluation metrics by 1 % accuracy, 4.7% precision, 6.6 % recall, and 0.7% F1-score.

Alzheimer's disease (AD) is the main cause of dementia and memory loss. Dementia has an important effect on the global healthcare system and society. Most cases of Alzheimer's disease occur in people around the age of sixty; however, it could be a result of concussions or other injuries to the brain. Alzheimer's disease requires a rapid diagnosis to begin effective treatment. The importance of early Alzheimer's disease diagnosis for therapeutic development and, finally effective patient care can really be affected. Current developments in deep learning provide promising findings for detecting different medical scans. This paper presented a CNN model for detecting Alzheimer's disease using MRI images. The suggested CNN model is extensively compared with several pretrained models, such as VGG16,VGG19,DenseNet201,ResNet101, Inceptionv3, MobileNet,Inception-ResNetV2, DenseNet169, and DenseNet121. Also, the evaluation of the proposed model is compared to relevant research efforts; findings indicate that it outperforms well, with an accuracy rate of 97%.

The garment sewing industry is a manufacturing sector dedicated to producing clothing and textile products. This industry encompasses a wide range of activities, including sewing and handling garments. These activities are considered value-added activities, while adjusting pieces, waiting for materials, and equipment maintenance are considered non-value-added activities. Tracking these activities is crucial for the production pipeline's efficiency. Typically, activity recognition involves the utilization of multiple deep learning models. Yet, the main obstacles in training these models revolve around identifying the most appropriate architecture and carefully fine-tuning numerous hyperparameters. This process is time-consuming and prone to errors. One potential solution to address these challenges is to incorporate ensemble learning by combining multiple, less powerful base models into a single, more robust model. Thus, this paper introduces an ensemble approach that integrates various activity recognition models through a combination of voting and meta-learning fusion techniques. Experimental results on baseline activity recognition models, including TwinsSVT, VideoMAE, and Conv3D, exhibit accuracy scores of 92%, 87%, and 70 %, respectively. Furthermore, a comparative analysis of different ensemble techniques is provided. The ensemble approach utilizing meta-learning outperforms the best baseline model, achieving an accuracy of 96% by employing extreme gradient boosting (XGBoost) as a meta-learner. This ensemble approach fuses three baseline models: TwinsSVT, VideoMAE, and Conv3D. By aggregating their predictions using various fusing methods including voting and meta-learners. The result of ensethis approach attains a 3.8 % increase in accuracy compared to solely relying on the highest performing individual model.

In the garment sewing industry, tracking worker activity is crucial for measuring production efficiency, particularly differentiating between value-added tasks (sewing, handling garments) and non-value-added activities (equipment maintenance, waiting). While traditional manual tracking methods struggle with scale and accuracy, recent efforts using deep learning, especially single-frame prediction, have aimed to address this challenge. However, these methods struggle to capture the flow of worker activity. By treating each frame independently, they miss the crucial element of time-dependent actions that define real-world work, leading to unreliable results. To overcome these limitations, this paper proposes an approach that considers both spatial (worker pose in a frame) and temporal information (sequence of worker actions) for robust worker activity recognition. The study compares the accuracy of different approaches in classifying tasks as value-added or non-value-added. This includes evaluating keypoint extraction methods (MoveNet, MediaPipe with LSTMs) against video classification using whole frames with convolutional neural networks (CNNs) including 3D convolutional layers (Conv3D), and video transformers (VideoMAE). Resulting in a 30% increase in accuracy when compared to keypoint extraction method.