بحوث سكوبس — م.د حسين عبد الأمير عباس الخميس

The disease course of Alzheimer’s disease (AD) results in severe brain degeneration which devastates both mental capacities and life quality. The proper detection of Alzheimer's disease at its early stage serves as an essential requirement for successful intervention strategies. Standard diagnostic procedures such as neuroimaging and cognitive screening methods require long duration and high costs but frequently reveal the disease after it reaches its later stages. The combination of artificial intelligence technologies and biosensors has developed into an effective method that enables real-time detection of Alzheimer's disease. AI models enable the analysis of complex biosensor data, resulting in the discovery of subtle biomarkers that signal early-stage AD, allowing for quick diagnosis and personalized medicinal regimens. Biosensor and AI development combine to create a system for immediate AD surveillance through trained device signals alongside EEG data, blood-related indicators, and gait pattern analysis. The comparison of two AD detection models uses support vector machines coupled with deep learning convolutional neural networks to establish their detection metrics which include accuracy levels sensitivity rates and precision measurements. Results show that the CNN approach reaches better accuracy rates compared to standard methods when classifying AD. The system design supports rapid non-invasive cost-effective screening procedures that are primed for implementation at large clinical settings. AI-powered biosensor integration enables early AD detection, opening new options for enhanced patient management and optimal intervention chances. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2026.

Respiratory ailments, particularly asthma, present significant healthcare challenges globally, necessitating prompt and precise identification for effective management. Conventional diagnostic techniques typically depend on clinical evaluations and pulmonary function assessments, which might be cumbersome and are often not available in isolated locations. This investigation delves into the use of artificial intelligence (AI) in monitoring respiratory health, with an emphasis on AI-enhanced asthma detection. By utilizing machine learning (ML) and deep learning (DL) methodologies, we scrutinize respiratory patterns, cough acoustics, and environmental influences to boost early diagnosis accuracy. The suggested approach incorporates convolutional neural networks (CNNs) and recurrent neural networks (RNNs) to process audio signals and time-series data gathered from wearable devices. A dataset containing both clinical and real-world respiratory sound recordings was employed to train and fine-tune these models. We applied feature extraction methods, including spectrogram analysis and statistical pattern recognition, to heighten classification precision. The experimental outcomes confirm that the AI model surpasses traditional diagnostic approaches in detection accuracy. Performance metrics, including sensitivity, specificity, and F1-score, substantiate the system’s capability to pinpoint asthma symptoms accurately. The results of this exploratory work illustrate the significant potential AI-driven systems hold for real-time asthma monitoring, facilitating timely interventions and decreasing the need for hospital visits. Incorporating AI with IoT-based wearable technology further bolsters accessibility and continuous monitoring capabilities. This study underscores the transformative impact of AI in respiratory health care, providing a scalable and economically viable option for asthma detection. Future initiatives aim to broaden the dataset and enhance model adaptability across varied patient demographics. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2026.

Machine learning methods, especially the K_Means clustering method, have demonstrated potential in analyzing medical data by facilitating pattern detection. However, the classic K_Means algorithm suffers from two major limitations: (1) its reliance on a single, often suboptimal distance metric (typically Euclidean), and (2) the lack of a mechanism to refine clusters post-assignment, which can lead to poor cohesion and misgrouping. To address these challenges, this paper proposes a novel enhanced K-Means clustering framework with two key innovations: (i) a hybrid distance approach that combines cosine and cityblock (Manhattan) metrics in a tunable weighted manner to better capture the structure of medical data and (ii) an efficient cluster refinement mechanism based on Z-score outlier detection to reassign distant samples and improve cluster quality. First, we evaluate K_Means using five distance metrics—Euclidean, cosine, cityblock, Chebyshev, and Minkowski—on two public medical datasets: Breast Cancer Wisconsin (BCW) and Heart Disease. Then, we introduce the hybrid distance strategy, systematically varying the weight between cosine and cityblock to identify the optimal combination. Following initial clustering, our refinement step identifies data points far from their cluster centroids (using Z-score) and reassigns them to more suitable clusters, significantly enhancing cluster homogeneity and separation. The proposed method is evaluated using multiple metrics: accuracy, precision, recall, F1-score, Adjusted Rand Index (ARI), homogeneity, and execution time. Results show substantial improvements over traditional approaches and advanced clustering methods (deep clustering and spectral clustering methods). For the BCW and Heart Disease datasets, the proposed method achieves accuracies of 0.9825 and 0.9000, outperforming Euclidean K-Means (0.8752, 0.8316) and cosine-based K-Means (0.9350, 0.8418). Homogeneity scores also enhance significantly from 0.7721 to 0.8676 (for BCW dataset) and from 0.4335 to 0.5352 (for Heart Disease dataset)—demonstrating the effectiveness of the refinement step. This work presents an original, practical enhancement to K_Means clustering for healthcare applications, offering improved accuracy, interpretability, and robustness through a hybrid distance strategy and a novel refinement mechanism. The results provide deeper insights into unsupervised learning for medical data analysis and support its potential in real-world clinical decision-making. © The Author(s) 2026.

The rapid evolution of intelligent systems, powered by artificial intelligence and machine learning, has created a fragmented research landscape. While numerous studies exist on specific applications, a holistic synthesis of their architectures, taxonomies, applications, and challenges is absent. This paper will bridge this gap by providing a comprehensive systematic review that integrates these disparate elements. This paper conducts a systematic review of over 100 peer-reviewed scientific publications, following a structured process to identify, analyze, and synthesize the current state of intelligent systems research. The review encompasses a wide range of domains, including healthcare, cybersecurity, data mining, and industrial automation. Our analysis yields a unified taxonomy and clarifies the core architectural components of intelligent systems. We identify and categorize key application domains and demonstrate their transformative impact. The review also synthesizes prevailing challenges, such as data quality, scalability, and ethical concerns, and pinpoints emerging trends, including the rise of multimodal AI and hybrid intelligent systems. To the best of our knowledge, this is the first review to offer a consolidated framework that integrates the architecture, taxonomy, applications, and cross-domain challenges of intelligent systems into a single reference. This work serves as a foundational guide for researchers and practitioners, facilitating future advancements in the development of efficient, scalable, and context-aware intelligent systems. © 2026 The Authors

Healthcare remains a critical focus due to its direct impact on human well-being. Diabetes, currently the fastest-growing chronic disease globally, poses severe health risks, including cardiovascular complications and kidney failure. Simultaneously, breast cancer has become the most prevalent cancer among women, particularly those in their 40s, surpassing other types. Early detection and diagnosis of these two diseases remain a substantial challenge, yet they are crucial for reducing mortality rates. Machine learning algorithms emerged as powerful tools in healthcare for disease classification and prediction, with the k-nearest neighbors (k-NN) being one of the most widely used supervised learning algorithm. Different traditional machine learning methods have been proposed, which are heavily specialized for specific datasets. More deeply, traditional k-NN relies on a static k-value, which may not provide optimal results across diverse datasets. This paper proposes a dynamic k-NN model that adjusts ‘k’ value based on local data characteristics, enhancing prediction accuracy. The proposed model is testing on two publicly available datasets; PIMA Diabetes and Breast Cancer Wisconsin (BCW) datasets. Our results are evaluated using different metrics that are; accuracy, precision, recall, F1_score, and execution time. The results of these metrics are as follows; (81.17%, 97.37%), (83.33% 100%), (54.55%, 86.05%), and (65.93%, 92.5%) for PIMA and BCW datasets respectively. These results demonstrate that the proposed model outperformed several state-of-the-art models. Thus, further highlighting its effectiveness and efficiency in medical data classification. © 2025

Accurate methods for early detection are required since heart disease is still a major problem in world health. In order to accurately forecast the occurrence of heart illness using electrocardiogram data, this research introduces a hybrid model called MLP-FRCNN (Multi-Layer Perceptron-Faster Region-Based Convolutional Neural Network). The suggested method uses the DNLMS algorithm to remove baseline fluctuations and motion artifacts from ECG signals before processing them. By utilizing Discrete Cosine Transform (DCT) and Fast Fourier Transform (FFT), we are able to extract features, with a particular emphasis on important components such the QRS complex. To improve the Faster R-CNN, the Honey Badger Algorithm (HBA) takes into account factors including computing efficiency and overlapped detecting boxes. Results from further tests show that, in comparison to contemporary methods, we achieve better accuracy, sensitivity, specificity, and F1 score. Machine learning, which began with data modification and accumulation, has evolved into a powerful tool for driving transformative change and remains a central component of the ongoing pursuit of artificial intelligence. Accurate detection and treatment for coronary heart disease patients are greatly enhanced by the suggested model’s higher speed of convergence and enhanced predictive capabilities. To improve accuracy, an FFNN combiner takes the estimates from both the Faster R-CNN and MLP and applies them to patient’s demographics information and low-order characteristics. With a 98% accuracy rate, the hybrid model outperforms both MLP (94% accuracy rate) and HBA-FRCNN (96% accuracy rate). © The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd. 2025.

The Internet of Things (IoT) refers to a network comprised of interconnected items, including computing devices and digital gadgets. Cloud-based IoT infrastructures are vulnerable to distributed denial of service (DDoS) attacks. A DDoS attack has the potential to incapacitate a server for an extended duration, resulting in service disruptions as a consequence of overwhelming system resources. This research presents a novel framework for mitigating DDoS attacks in IoT networks. The proposed system leverages the fog-cloud architecture to provide efficient, lightweight, and precise attack mitigation. Notably, the mitigation process is executed at the fog layer. The suggested fog layer uses Particle Swarm Optimization (PSO) to make allocating resources easier, which makes it possible for the mitigation framework to be set up quickly. This approach addresses the challenges associated with resource management on resource-constrained IoT devices. The mitigation framework uses the Fitness Leader Optimization (FLO) approach to construct a trained database, taking into consideration factors such as the needed time, the size of the request, and the number of created requests. The FLO system employs multilayer perceptron (MLP), k-nearest neighbors (KNN), and support vector machine (SVM) classification algorithms to effectively mitigate the assault. The results of this study show that adding classification algorithms to our framework made it easier to test networks for Internet of Things (IoT) devices, especially when the Particle Swarm Optimization (PSO) method was used together. The mitigation framework demonstrates a minimized fitness value of 0.284556 seconds, showcasing enhanced resource utilization and processing time optimization for IoT nodes and servers in a distributed fog environment. The total average of resource utilization is improved to 6.0850%, processing time is decreased to 17.0397, and fitness value is decreased to 0.0258 seconds in the proposed DDoS attack mitigation system. The machine learning classification model achieves high accuracy, with SVM leading at 99.6785% compared to others, emphasizing the robustness of the proposed framework in securing IoT networks. © 2025 University of Baghdad-College of Science. All rights reserved.

This research investigates the effectiveness of a CNN-LSTM hybrid model for forex investment risk prediction, demonstrating superior accuracy and robustness compared to traditional models. The proposed model achieves an F1-score of 0.94, outperforming standalone LSTM (0.88) and statistical approaches like ARIMA (0.67) and GARCH (0.72). Incorporating the Directional Change Framework enhances trend alignment, reaching 94.5% for GBP/USD and 96.2% for EUR/USD, compared to 81.3% and 84.7% without it. The model significantly reduces forecasting errors, achieving MSE 1.7) further validates its effectiveness in risk-adjusted forex forecasting. Short-term predictions (1-hour and 4-hour) dominate usage (72.4%), while long-term forecasting (30-day, 90-day) remains limited (3.2%). With a 57.6% adoption rate, CNN-LSTM outperforms Transformer-based methods (22.3%) and GRU (10.4%) in AI-driven forex modeling. Although its predictive effectiveness, computational requirements are still a constraint since Transformer-based models consume 10x more processing. These results underscore the promise of deep learning models in forex investment risk forecasting, maximizing financial decision-making through enhanced accuracy and real-time responsiveness. © 2025 IEEE.

Augmented Reality (AR) systems are increasingly integrating artistic visual effects to enhance user experience, with Neural Style Transfer (NST) offering a powerful means of transforming visual content. However, existing NST methods often struggle with maintaining real-time performance and stylistic richness on resource-constrained AR platforms. Current approaches usually lack temporal coherence and produce either overly smoothed or computationally intensive outputs, rendering them unsuitable for dynamic, real-time AR applications. To address these challenges, this study proposes a novel framework, Deep Dream-Guided Neural Style Transfer (DD-GNST), which integrates Deep Dream's feature amplification capabilities with efficient Neural Style Transfer (NST) for enhanced artistic expression. The proposed DDGNST framework employs a lightweight encoder-decoder structure, where Deep Dream layers guide stylization by amplifying specific neural activations, yielding more vivid and surreal artistic effects. This enables the system to generate temporally stable, visually appealing output in real-time, suitable for AR headsets and mobile devices. Experimental results demonstrate that DD-GNST outperforms conventional NST models in both processing speed and visual quality, maintaining temporal consistency across frames. This approach paves the way for immersive, artistic AR experiences in live video streaming, gaming, and creative mobile applications. The proposed method achieves a frame rate increase of 98.4%, a temporal consistency score of 97.8%, a stylization quality of 96.3%, and a model size reduction of 98.5%. © 2025 IEEE.

The nature of images can differ in texture, contrast, illumination, noise levels, and structural patterns. The descriptor suitable for one image may not be optimal for another. Therefore, this paper proposes a new hybrid diagnostic model that combines multi-descriptor feature extraction with a Deep Belief Network. It is used to classify Acute Lymphoblastic Leukaemia. The proposed model consists of two phases: feature extraction and classification. Three descriptors, Histogram of Oriented Gradients, Scale-Invariant Feature Transform, and Convolutional Neural Network are employed in the feature extraction phase. Each descriptor captures different aspects of the image using distinct computational techniques. The Deep Belief Network was trained on each group of features individually. Three trained Deep Belief Network were produced with each data extract by descriptors. The membership function between the training set and the test data determines which DBN will be selected. The model was tested and evaluated on the 10,661 Leukaemia images of the C-NMC_Leukaemia dataset. It consists of two classes of images: 7272 images of Leukaemia cancer and 3389 of the Benign. Experimental results showed that the proposed model achieved an accuracy outperforming several recent methods. The accuracy of the proposed model reaches 96.87%, while the best accuracy of the recent works is 94.91%. © 2025, American Scientific Publishing Group (ASPG). All rights reserved.

In today’s digital era, most electrical gadgets have become smart, and the great majority of them can connect to the internet. The Internet of Things (IoT) refers to a network comprised of interconnected items. Cloud-based IoT infrastructures are vulnerable to Distributed Denial of Service (DDoS) attacks. Despite the fact that these devices may be accessed from anywhere, they are vulnerable to assault and compromise. DDoS attacks pose a significant threat to network security and operational integrity. DDoS assault in which infected botnets of networks hit the victim’s PC from several systems across the internet, is one of the most popular. In this paper, three prominent datasets: UNSW-NB 15, UNSW-2018 IoT Botnet and recent Edge IIoT are using in an Anomaly-based Intrusion Detection system(AIDS) to detect and mitigate DDoS attacks. AIDS employ machine learning methods and Deep Learning (DL) for attack mitigation. The suggested work employed different types of machine learning and Deep Learning (DL): Random Forest (RF), Support Vector Machine (SVM), Logistic Regression, and Multi-layer perceptron (MLP), deep Artificial Neural Network (ANN), and Long Term Short Memory (LSTM) methods to identify DDoS attacks. Both of these methods are contrasted by the fact that the database stores the trained signatures. As a results, RF shows a promising performance with 100% accuracy and a minimum false positive on testing both datasets UNSW-NB 15 and UNSW-2018 Botnet. In addition, the results for a realistic Edge IIoT dataset show a good performance in accuracy for RF 98.79% and for deep learning LSTM with 99.36% in minimum time compared with other results for multi-class detection. © IBERAMIA and the authors.

Spam emails are unsolicited, unwanted emails that are usually sent in large quantities by advertisers and scammers. They are often sent for the purpose of promoting a product or service or for phishing, which is the attempt to obtain sensitive information such as usernames, passwords, and credit card details by pretending to be a trustworthy entity in an electronic communication. Deep learning algorithms can be used to identify spam emails by analyzing large datasets of email messages and learning to recognize patterns and trends that are indicative of spam. For example, a deep learning algorithm could be trained on a dataset of spam and non-spam emails and then be able to identify spam emails with a high degree of accuracy based on the patterns and trends it has learned from the training data. For the current work, machine learning by using the random tree is used to determine the best features with the leading deep learning hybrid Deep Neural Network Convolution Neural Network (DNN-CNN) techniques in the field of disclosure of incidental messages (spam and non-spam). The results showed that a high accuracy rate (99.8%) was obtained comparing with minimum false positive rate to the other works. Copyright: © 2024 The authors.

Neural networks are effectively used in a variety of applications including data mining. The neural network can realize different complex nonlinear functions by making them attractive to identify a system. One of the most important issues of classifying datasets through neural networks is the formation of an ideal network, that consists of many successive steps like set parameters. Perhaps the most prominent parameter is the learning rate. Indeed, choosing an appropriate learning rate value is one of the things that greatly helps to control the overall network performance. In contrast, any inappropriate value for the learning rate negatively affects the classification model and can therefore destabilize the model’s performance and thus seriously deteriorate its quality. This paper presents a new model by adopting a cyclical learning rate instead of using a constant value for training deep neural networks by Multi-Layer Perceptron (MLP) architecture. This model is tested on various real-world datasets; Electricity, NSL- KDD, and four sub-datasets of HuGaDB (HuGaDB-01-01, HuGaDB-05- 12, HuGaDB-13-11, and HuGaDB-14-05). The proposed model achieves an accuracy of, 89.57%, 99.12%, 99.2%, 97.83%, 96.19%, and 99.85% for these datasets respectively. Accordingly, the proposed model outperforms many previous models. As a result, the deep neural network models can be more effective when they adopt an appropriate value for the learning rate. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2024.