Journal of Al-Qadisiyah for Computer Science and Mathematics

With the increasing complexity of cybersecurity threats, Network Intrusion Detection Systems (NIDS) have become essential tools for securing organization networks. These Systems are designed to monitor traffic in real-time and detect unauthorized or malicious activities. Traditional machine learning algorithms have been extensively used for intrusion detection; however, most rely on shallow learning techniques, which are often ineffective in handling high-dimensional and complex network data. This study proposes a deep learning-based intrusion detection framework to address these limitations. The proposed method employs a Deep Belief Network (DBN) for deep feature extraction and dimensionality reduction, followed by a Multi-layer Perceptron (MLP) trained using the backpropagation algorithm to classify and detect intrusions. The approach is evaluated using two benchmark datasets: KDD CUP 1999 and NSL-KDD, selected for their diversity, labeled attack categories, and widespread used IDS performance benchmarking. Experimental results demonstrate that the proposed DBN-BP model achieves an average accuracy of 98.19% on KDD CUP 1999 and 98.17% on NSL-KDD. Experimental results demonstrate that the DBN-MLP approach achieves a recognition rate improvement of 13.45% over traditional SVM-based classifiers, additionally, on the KDD CUP 99 dataset, the DBN-MLP model demonstrates a 12.46% improvement over Decision Tree classifiers. These results confirm the model's superior learning capacity and enhanced ability to generalize to previously unseen attack types. This can be attributed to the hierarchical feature extraction capabilities of the Deep Belief Network combined with the classification strength of the Multi-layer Perceptron. Given these improvements, the DBN-MLP approach is highly suitable for real-time Network Intrusion Detection System (NIDS), especially in enterprise-level and cloud-based environments where high detection accuracy and responsiveness are critical.

Quick Internet of Things (IoT) and Wireless Sensor Networks (WSN) proliferation have considerably raised real-life and automation monitoring, data-based decision-making over various fields such as industrial systems, healthcare, and smart cities. Although great IoT device development defines security vulnerabilities and operational risks, it signifies strong anomaly diagnosis algorithms for recognizing system failures, unusual behaviors, and cyber threats. Traditional rule-based and statistical techniques cope with controlling active, massive, and high-dimensional IoT data aspects, creating methods of machine learning (ML) that are promising alternatives for appropriate and scalable unusual diagnosis. The present paper shows a general review of ML-based unusual diagnosis strategies in IoT and WSN, grouping them into hybrid, unsupervised, and supervised learning methods. In addition, it examines deep learning architectures, such as Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM), and Transformer-based models, highlighting their strengths incapturing complex spatial and temporal dependencies in sensor data.Despite their efficiency, ML-based techniques meet some issues like real-life limitations, data scarcity, high computational costs, adversarial vulnerabilities, and a shortage of generalization over various IoT areas. For considering such issues, this review describes the present paper's directions. Though state-of-the-art methods’ analysis and highlighting future trends, the present paper targets present worthy perspectives for investigators and practitioners in improving more adaptive, safe, effective ML-based unusual diagnosis responses for IoT and WSN

Breast cancer is a life-threatening disease for women, and early detection of this disease will lead to a longer lifespan. Breast cancer is a leading cause of death among women worldwide. In this study, we designed a proposed convolutional neural network (CNN) model that consisting of 39 layers. The model begins with feature extraction from input mammogram images using a series of convolutional layers, batch normalization, ReLU activation functions, and max-pooling layers.This is followed by six fully connected layers for classification. Overall, the proposed model includes a three-layer feature extractor and a two-layer decision-making module designed specifically for breast cancer detection. The model achieved impressive performance metrics, with an accuracy of 98%, precision of 95%, recall of 96%, and an F-score of 96.4% if number of epoch equal to 5 and 100% if epoch equal to 10.

Parkinson’s disease (PD) progresses through the nervous system as a neurodegenerative condition which causes severe effects on motor function and generates life quality problems. Standard diagnostic approaches face two main limitations that include a subjective basis and prolonged diagnostic delays. This research develops an AI-driven diagnosis classification system which involves analyzing datasets from public domains together with medical patient information stored at Baghdad hospitals. The decision tree and random forest models together with SVM and CNN provided implementation for PD diagnosis assessment through performance metrics testing. The CNN model achieved the highest correctness rate of 94.3% as well as precision of 93.1% and F1-score of 93.5% to outperform the other proposed models during testing. The application of CNN on the local dataset achieved a strong 90.7% accuracy because it successfully adapted to different data quality and format conditions. The research determined that tremor frequencyand voice pitch variation together with hand movement speed proved to be the most effective diagnostic features. Research demonstrates AI diagnostic technologies can assist early Parkinson disease identification at healthcare sites which lack resources byshowing the necessity of better digital networks and standardized research data alongside professional education for medical staff.

One of thecharacteristics of 5G networks is their ability to create multiple virtual networks across a shared infrastructure, with resources dynamically allocated to meet the needs of different applications. However, managing these partitions in real-time remains achallenge due to their dynamic and heterogeneous nature. To overcome this challenge, this paper proposes an incremental learning model (ILM) that gradually learns from changing network data, seeking to improve the accuracy of partition selection. Unlike traditional models trained from static datasets, the ILM continuously updates its knowledge without the need to retrain from scratch. The results demonstrate the effectiveness of the proposed method, with an accuracy rate of 94.7%, while also demonstrating the ability to train using new traffic patterns and environments.

This paper proposes a hybrid machine learning-based framework for Software Defined Networking (SDN) environments, integrating a Deep Q-Network (DQN) for intelligent routing optimization and an Autoencoder for anomaly detection. The system dynamically learns optimal routing policies while simultaneously identifying network threats in real-time. Both real and synthetic datasets were used to validate the framework, demonstrating improved network efficiency and detection accuracy. Experimental results confirm the framework’s capability to adapt to diverse traffic patterns, optimize network flow, and secure SDN infrastructures effectively.

Schizophrenia (SZ) is a chronic and severe mental disorder characterized by impairments in cognitive skills, perceptions, emotions, and social interactions. A timely and accurate diagnosis is crucial for improving prognosis anddeveloping effective treatment strategies. Recently, researchers have utilized computational models to enhance the effectiveness and speed of schizophrenia diagnosis using electroencephalogram (EEG), consequently reducing clinical workload. This research investigates the integration of traditional signal processing techniques, feature extraction methods, and artificial intelligence (AI), including machine learning (ML) and deep learning (DL), for the categorization of schizophrenia (SZ) utilizing EEG data.The electroencephalogram, a crucial tool for assessing cerebral activity, has demonstrated importance in mental health research. Upon acquiring brain data, various signal-processing techniques are employed to extract pertinent information from the temporal, frequency, and spatial domains. The gathered properties, encompassing mean, variance, and band power, are the basis for recognizing EEG signals.Traditional machine learning techniques, such as Decision Trees and Support Vector Machines (SVMs), provide interpretability and effectiveness with constrained datasets.In contrast, deep learning techniques, such as convolutional neural networks (CNNs) and extendedshort-term memory networks (LSTMs), excel in analyzing complex EEG patterns; however, they require extensive data and significant computational resources. The study examines the challenges associated with implementingAI in the diagnosis ofschizophrenia, including ethical concerns and issues with data quality. These difficulties require collaborativeand ethically sound approaches to ensure reliable advancement in the area.The research highlights the importance of employing many approaches to improve diagnosis accuracy, showcasing the potential of AI-driven solutions in the classification of schizophrenia. This review offers a comprehensive examination of contemporary literature, encompassing themes, approaches, and conclusions. The aim is to identify significantadvancements and provide insights that helpresearchers and cliniciansunderstandand tackleschizophrenia throughinnovative AI-driven approaches.

Recently, there has been a notable increase in the need for the generation, dissemination, and storage of extensive volumes of multimedia data, especially digital images,generated by different intelligent devices and sensors. Along with other security vulnerabilities, such activity results in unauthorized access and false use of data. Embedding a watermark design into a digital cover and then removing it helps to address ownership conflicts and copyright infringement concerns related to the media data. Deep-learning methods are currently rather helpful for watermarking because of their high accuracy, great precision, and strong learning ability. This article presents a comprehensive examination of watermarking techniques used in deep learning contexts. We start by explaining the basic ideas of both traditional and learning-based digital watermarking, and then we look at common watermarking methods that use deep learning models. We then provide a succinct summary and comparison of the most recent contributions in the literature. Finally, we highlight the challenges related to obfuscation and suggest avenues for further study.

Alzheimer's disease is a progressive neurodegenerative disorder that primarily affects individuals aged 65 and older, leading to irreversible memory loss and cognitive decline. Earlydetection plays a critical role in managing the disease and improving patient outcomes. In recent years, numerous studies have investigated the development of automated systems to identify the stages of Alzheimer's disease using advanced deep learning methods. This paper provides a structured literature review focused on the use of Vision Transformers (ViTs) and metaheuristic optimization algorithms for early diagnosis. The reviewed studies demonstrate that ViT-based models outperform traditional approaches in extracting spatial and temporal features from brain imaging data, achieving classification accuracies exceeding 96% on widely used datasets such as ADNI and OASIS. Additionally, the review addresses key challenges in processing 3D medical images and highlights ongoing efforts to develop hybrid architectures that integrate the strengths of Convolutional Neural Networks (CNNs) and Vision Transformers (ViT). The paper also explores how collaborative learning strategies can enhance model training while preserving patient privacy, making these techniques more suitable for real-world clinical applications.

As technology has advanced, data protection has gotten more challenging. Information security now depends on cyber security. These days, protecting information is one of the biggest problems. Unquestionably, the Internet has created a new avenue for exploitation known as cybercrime because of the limitless amount of free websites. To combat these cybercrimes, both the public and business sectors are implementing a number of initiatives. Managing cyber security is a major challenge. In addition to presentingnew trends and issues in the realm of cyber security, this research study focuses on upcoming technologies. It is anticipated that the thorough review study offered to scholars studying IT and cyber security will be beneficial. This analysis examines current advancements, difficulties, and new technology in the cybersecurity space. We concentrate on important ideas like cyber threats, and cyber threat intelligence. The relevant literature from 2020 to 2024 is evaluated using a systematic review methodology. This study aims to explore current trends, challenges, and future directions in cybersecurity and identify major cyber threats

Face recognition technology has significantly advanced and become essential in various applications, including security systems, human-computer interaction, and biometric authentication. This work introduces an effective face Recognition for recognizing a person from facial images. This work utilizes state-of-the-art preprocessing and augmentation techniques, histogram equalization, Gaussian blur, and Canny edge detection in enhancing quality and diversity within the data. These will ensure that all facial features are captured, including those extracted by DeepFace, HOG, LBP, and Gabor filters. Some of the machine learning methods applied and tested on this CAS-PEAL dataset include XGBoost, Decision Tree, and Random Forest. Of these, the best performance was by XGBoost, at near perfection. The outcomes hereby show that an efficient fusion of multi-feature extraction approaches with the recent machine learning methods improves the reliability of face recognition against every change in environmental conditions.

Multimedia video applications significantly impact video quality prediction, widely regarded as one of the most challenging problems. The Quality of Experience (QoE) prediction of the video mimics the satisfaction of the content of the video as humans perceive it. Machine learning and deep learning models have applied numerous methods to obtain QoE predictions. Some of these methods are full reference or reduced reference (half reference); others are no reference. In this paper, we attempt to explore,evaluate, and analyzethe different scenarios and models related to QoE predictions for videos using deep learning. We have conducted a comprehensive examination to address the limitations of the existing models. Moreover, we suggest a new framework to overcome the limitations of the existing models

Vehicle classification and tracking are fundamental components of intelligent transportation systems (ITS), enabling real-time traffic monitoring, surveillance, and autonomous driving. This paper provides a comprehensive review of modern deep learning and artificial intelligence methods used in vehicle classification and tracking. It highlights key methods such as CNNs, YOLO, Faster R-CNN, Kalman Filters, DeepSORT, and LSTM networks, with a particular focus on hybrid CNN-LSTM models that combine spatial and temporal features for robust performance in dynamic scenes.The review identifies major challenges reported in the literature, including occlusion, adverse weather conditions, real-time processing requirements, and data privacy concerns. It also outlines the most common application scenarios, such as smart surveillance, urban traffic control, and autonomous navigation.Based on the current trends, the paper recommends future directions involving vision transformers, reinforcement learning, edge computing, and multimodal sensor fusion. The goal is to offer researchers and practitioners a structured overview of the state-of-the-art, while highlighting opportunities for improving adaptability, scalability, and efficiency in smart transportation systems.

This articlereviewed some of the roles played by artificial intelligence (AI) as a modern technology in the educational system. We also tested an AI-based application to detect cheating during exams by analyzing facial and eye movements. The system uses computer vision techniques via the OpenCV library, along with an alert device designed based on Arduino, allowing for real-time monitoring and response. Haar Cascade algorithms were used to detect faces and eyes, and head movement and gaze direction were tracked using optical flow and motion center calculation techniques. An SVM classification algorithm was used to identify suspicious behavior based on features extracted from the image.Experiments showed that the proposed system achieved an accuracy of more than 90% indetecting cheating cases in real time. There are also some challenges such as poor lighting. The research is considered a contributing factor in enhancing academic integrity in educational institutions. The results indicate that AI-based surveillance systems can be effective in educational settings, and suggest future development of the system using deep learning techniques and multiple cameras to improve performance.

Otitis media (OM) is a common inflammatory condition, particularly in children, and poses significant diagnostic and treatment challenges. It also significantly affects adults, especially those with atopic conditions. OM can lead to complications such as hearing loss and speech delays. The advent of electronic health records, big data, and artificial intelligence (AI) offers transformative opportunities in OM diagnosis. AI plays a crucial role in enabling early detection and enhancing diagnostic precision through advanced image analysis and predictive modeling. This review paper explores modern techniques used for the diagnosis of OM, with an emphasis on both traditional and machine learning approaches. A wide range of studies has been evaluated, demonstrating the application of AI in improving diagnostic accuracy and treatment planning. Notably, AI approaches—particularly deep neural networks—have shown remarkable success in otoscopy image analysis. Additionally, recently developed hybrid models that combine multiple techniques have outperformed individual approaches. Despite these advancements, challenges remain, including limited dataset standardization and issues with image quality

This study explores the application of neural networks in financial data analysis and corporate performance evaluation, addressing challenges of nonlinearity and data noise. It evaluates multiple architectures (feedforward, recurrent, and convolutional), supported by real-world Python implementations and performance metrics such as accuracy and AUC. Key limitations like interpretability and overfitting are discussed, with proposed remediesand future research directions.

This study presents an enhanced Speech Emotion Recognition (SER) model that integrates Convolutional Neural Networks (CNN) with the Bat Algorithm (BA), a nature-inspired metaheuristic optimization technique. The objective is to improve the accuracy and generalizability of emotion classification from speech by optimizing the neural network architecture. The model utilizes handcrafted acoustic features—pitch, energy, zero-crossing rate (ZCR), and Mel-Frequency Cepstral Coefficients (MFCCs)—as inputs. Before being input into a deep neural network whose hyperparameters are modified using the Bat Algorithm, these characteristics are preprocessed and normalized. The last model uses a Gamma Classifier (GC) and Error Correcting Output Codes (ECOC) to guarantee strong categorization. Experimental findings employing benchmark datasets like IEMOCAP, EMO-DB, and Berlin DB show better performance, with validation accuracy climbing as high as 98.44%. This hybrid design offers a consistent method for practical emotion identification systems and outperforms traditional methods.

The great development in information technology and the Internet of Things is accompanied by several concerns, the most important of which are security vulnerabilities and cyber-attacks that affect the joints of the connection between resources, as old methods are no longer sufficient to reduce security vulnerabilities and protect interconnected systems.The study aimed to design a strategy that works on defense, combating deception, controlling operating systems, and diversity in the process of operational mobility with a process of updating and evaluating in real time by deploying deceptive nodes and dynamically moving between operating systems of Internet of Things devices and continuously monitoring the attack surface.This study focuses on the security returns resulting from the designed system, as it adopts a multi-layered security approach to protect Internet of Things networks from complex and targeted threats. This defensive approach is a powerful approach to increase the complexityof attackers and try to stop them and improve the overall resilience of the Internet of Things ecosystem. The effectiveness of the proposed defense-in-depth strategy is evaluated through simulation and performance analysis, which indicates its ability to mitigate various types of attacks while maintaining acceptable overhead.

This survey provides a systematic review of IoT security by conducting a strict examination of over 200 research papers published during 2014-2023 and ultimately shortlisting the most significant contributions based on citation importance and relevance. The major findings show that conventional security solutions are increasingly inadequate to address the specific challenges of IoT environments predicted to connect 84 billion devices by 2025. Our research identifies critical security vulnerabilities at the perception, network, and application layers, while examining the prospects of new solutions including blockchain, artificial intelligence, and edge computing solutions. The integration of these technologies shows promising results in strengthening IoT security, even though deployment challenges remain in resource-constrained environments. Domain-specific findings depict the need for tailored security frameworks for industrial, healthcare, and home automation systems. This survey concludes that future IoT security requires scalable, autonomous, and adaptive defense mechanisms that can overcome exponentially increasing threats while providing interoperability on heterogeneous platforms

The widespread use of instant messaging applications, such as WhatsApp, has converted these platforms into dynamic venues for social interaction. The rising prevalence of commercial advertisements may negatively affect the quality of social connections. This study investigates the impact of ads on social interactions in WhatsApp groups using the Iraqi dialect, applying natural language processing and artificial intelligence techniques. We propose a methodology employing the BERT model to classify WhatsApp messages in the Iraqi dialect into three primary categories: advertisements, social discourse, and neutral communications. The primary objective is to assess the impact of advertisements on the dynamics of discussions among users in groups. A dataset comprising 5,000 messages was meticulously gathered and categorizedinto two classifications: advertisements and social discussions. The pre-trained CAMeLBERT model underwent fine-tuning on this dataset by incorporating a classification head and training for 50epochs with a batch size of 8 and a learning rate 2e-5.Experimental results indicate that the model attained an F1-score of 97%, effectively differentiating between commercial communications and casual conversations. Approximately 35% of the messages wereclassified as promotional content. A conventional SVM model utilising TF-IDF features was implemented to assess performance, attaining merely 81.3% accuracy, underscoring the superiority of the transformer-based methodology. These findings indicate that the increasing prevalence of advertisements may discreetly disrupt the natural flow of conversations in digital communities, necessitating the implementation of sophisticated filtering systems.

Heart disease, which frequently results from blockage of the coronary arteries, the blood channels that supply the heart with oxygen-rich blood, is still one of the top causes of death globally. Plaque and fatty deposits that accumulate along the arterial walls are the main causes of this blockage, which makes the arteries narrow and limits blood flow. Even with cardiac disorders' seriousness and potentially fatal consequences, early detection remains a significant challenge in the medical field, oftendue to the complex and subtle nature of early symptoms. This diagnostic difficulty highlights the need for advanced computational tools that can support clinical decision-making. In this context, this study investigates the application of machine learningalgorithms to heart disease risk prediction. Five artificial intelligence models—Support Vector Machine (SVM), Random Forest (RF), K-Nearest Neighbors (KNN), Decision Tree (DT), and Naive Bayes (NB)—were tested on a dataset of 1,888 records and 14 attributes. In order to increase data quality and guarantee that models can learn patterns efficiently, preprocessing was used to clean and prepare the data first. This was followed by hyper parameter optimization for the SVM and KNN models. The aim of hyper parameter optimization is to maximize the model's performance on the data. With an accuracy of 96.30%, Random Forest outperformed the other models under evaluation. Decision Tree came in second with 95.59% , SVM with 94.55% ,KNN with 0.9425 and Naive Bayes with 0.6966 . These studies demonstrate how machine learning may be used to identify cardiac illness early on by identifying intricate patterns in data and providing more precise results than conventional techniques.

The purpose of malware design is to attack computers specifically for data theft operations and system network disruption to gain access to sensitive data. The mobile platform risk increases due to the widespread Android operating system use because its design keeps it open-source by nature. The research demonstrates the use of deep learning composite models for the detection of Android software malware. The proposed method leverages a pre-trained GoogleNet convolutional neural network for feature extraction from malware-related data and employs a deep recurrent BiLSTM network for classification. The MRMR (Minimum Redundancy Maximum Relevance) algorithm enables feature selection in a process that optimizes model classification speed as well as accuracy levels. This detection system addresses three main malware detection obstacles: the analysis of extensive data sets as well as the quick development of complex malware code and its well-camouflaged signatures. The simulation results showed that the model successfully detected various malware types with 99.30% success rate. Research results show that contemporary security threats need complex malware detection systems because conventional antivirus solutions have proven ineffective against new security threats. The constructed AI-based malware detection framework serves as an advanced solution that delivers improved protection for Android devices and their cyber threats.

This paper presents a survey of studies and research on machine learning techniques, deep learning, as well as research using optimization algorithms, and evolutionary algorithms, in relation to discoveries of diagnosing thyroid disorders. The paper includes an analysis of recent studies. The techniques of researchers in this field of diseases (thyroid diseases) are shown. As well as the results of the aforementioned studies on accuracy factors, Precision, Recall, F1-scor and. Advantages and disadvantagesof advanced algorithms. By analyzing the current methods that were surveyed, it was proven that transfer learning techniques, as well as techniques that use optimization algorithms, are the most efficient. In some research, preprocessing plays a major role in obtaining better results in the model training stage

Precise and stable trajectory tracking in mobile robotics is challenging due to dynamic disturbances and uncertain conditions. Traditional controllers like P-D and PID often underperform in non-linear, unpredictable environments. They also lack the adaptability needed for real-time, vision-based control systems. To address these issues, this paper explores and compares the effectiveness of three control approaches: P-D, PID, and a Fractional Order PID (FOPID) controller optimized using the World Cup Optimization Algorithm (WCOA). The control system is driven by inputs from a computer vision setup, and the controllers are evaluated on their ability to maintain accurate setpoint tracking under simulated dynamic disturbances. The FOPID controller’s parameters are tuned using WCOA, enhancing its adaptability and precision. The robustness of WCOA is also validated using standard benchmark functions including Sphere, Rastrigin, Ackley, Rosenbrock, and Griewank. Simulation results show that the WCOA-tuned FOPID controller significantly outperforms both P-D and PID controllers in minimizing error, maintaining control stability, and ensuring accurate trajectory tracking. Its superiority is also demonstrated through real-time tests on a mobile robot. Furthermore, feasibility evaluations reveal high levels of usability, operability, and learnability. These results highlight the strong potential of WCOA as a robust optimization method for advanced control systems and provide a foundation for future research in dynamic environments, hybrid optimization techniques, and broader engineering applications

Key frame (KF) extraction process is considered as a crucial task in video structure analysis, it plays important role in video summarization, content analysis, video compression and so on. This process aimed to give a good video summarization by extracting the frame or set of frames that provide a comprehensive representation for the video sequence and removing the other frames that are considered redundant. In this paper,a new method has been proposed, the proposed method utilized Histogram of Oriented Gradient (HOG) as feature and adaptive thresholding technique, enabling the detection of substantial changes in visual content between consecutive frames in each video shot. By calculating the HOG differences and applying adaptive threshold, the method divides the shot's frames into groups, then from each groups a key frame with substantial features is selected, while other frames are considered as redundant and removed. Experimental evaluations on different videos, shows that the proposed methods able to extract key frames that accurately reflect the video's primary content and produce a good summarization. The results reducethe redundancy while preserving the essential idea of the video.

In this article, we investigate the adjacency matrix and the eigenvalues of the Ideal-based-zerodivisor graph for a given ideal of the finite commutative ring . Additionally, we define the notion of projection graph of the Idealbased-zerodivisor graph as a graph with vertices for all and edge connecting with if every element in is connected with . We consider the ring for some special cases of . Ifwhere (where ) and is non-prime ideal of , then we determined the determinant of the adjacency matrices of the Idealbased-zerodivisor graph of and satisfies (())We showed that in the Ideal-based-zerodivisor graph (), where ∏, is not prime ideal of and not zero ideal of , then |()|∏and whenever ∏then ()().Finally, we investigate some basic properties of the edge ideal of the graph . We calculated the graph's depth, girth, diameter, Betti number, regularity, and projective dimension.