Mesopotamian Journal of Computer Science

To incite modern day crop production and ensure sustainability, exact crop recommendations are key to
the process. This study pays significant attention to the need for the use of big data tools in studies
involving comprehensive data sets that contain information on soil and other environmental
characteristics. The set of data used in this research includes Nitrogen, Phosphorus, and Potassium
content coordinated with Temperature, Humidity, pH Value, and Rainfall. Knowing these factors is to
make a favorable decision about improving agricultural products yield, availability and management of
the resources, as well as general well-being of the crops. Specialized advisory on crop type according to
the specific data of the soil is likely to click the concept of yield by dealing with nutrient gaps and variable
environments. This approach also guarantees reasonable distribution of the limited resources as well as
cutting the cost of production and the impact on the environment through preserving the soils and keeping
off the faster pace of nutrient leaching. This method of producing plants also helps in minimizing the
utilization of fertilizers and enhanced inputs hence minimizing the possible negative impacts on the
environment as well as increasing the lifespan of the field. Furthermore, specific advice also assists the
farmers in handling changes in climate patterns in a way that increases crop’s ability to withstand the
impacts of more frequent episodes of the extremities and manage the risks implicated in those
circumstances. Since climate issue plays significant role in threatening the global food security, the
strategic maneuvering capacity which enables to analyze the environment and respond to the changes in
a timely manner has gained added importance. This also eliminates the losses likely to be incurred as a
result of adverse weather factors through the use of data and empirical analysis, which provides
management measures to be adopted in order to protect crops and ensure steady food production. In the
end, enhanced decision making regarding the types of crop that do well on the soil can easily enhance
farmers’ revenues via; improved yields, improved crop quality, lower input expenses etc. This does not
only positively affect the economics of the farming enterprises but it is also a way of supporting the
stability of agricultural businesses globally. This has informed a build of an Artificial Neural Network
(ANN) model which is capable of analyzing the large and rich dataset of soil information to specify
accurate crop recommendation. In the validation of the model, this achieved an accuracy level of 99% as
it manifested its ability to provide succor in decision making analysis in the sector of agriculture. In
comparison to the conventional algorithms in Machine Learning like Logistic Regression, K-Nearest
Neighbors (KNN), Multinomial Naive Bayes algorithms, our ANN model exhibits higher accuracy and
consistent performance. This study thus establishes the importance of adopting the higher inclusive
machine learning with improved data about soil to enhance a more efficient, resilient, and sustainable
agriculture era.

Wetlands are essential ecosystems that provide ecological, hydrological, and economic benefits.
However, human activities and climate change are degrading their health and jeopardizing their longterm sustainability. To address these challenges, the Internet of Wetland Things (IoWT) has emerged as
an innovative framework integrating advanced sensing, data collection, and communication technologies
to monitor and manage wetland ecosystems. Despite its potential, the IoWT faces substantial security
and privacy risks, compromising its effectiveness and hindering adoption. This survey explores
integrating machine learning (ML) and deep learning (DL) techniques as solutions to address the security
threats, vulnerabilities, and challenges inherent in IoWT ecosystems. The survey examines findings from
231 sources, encompassing peer-reviewed journal articles, conference papers, books, book chapters, and
websites published between 2020 and 2025. It consolidates insights from prominent platforms such as
the Springer Nature, Emerald Insight, ACM Digital Library, Frontiers, Wiley Online Library, SAGE,
Taylor & Francis, IGI Global, Springer, ScienceDirect, MDPI, IEEE Xplore Digital Library, and Google
Scholar. Machine learning and DL methods have proven highly effective in detecting adversarial attacks,
identifying anomalies, recognizing intrusions, and uncovering man-in-the-middle attacks, which are
crucial in securing systems. These techniques also focus on detecting phishing, malware, and DoS/DDoS
attacks and identifying insider and advanced persistent threats. They help detect botnet attacks and
counteract jamming and spoofing efforts, ensuring comprehensive protection against a wide range of
cyber threats. The survey examines case studies and the unique requirements and constraints of IoWT
systems, such as limited energy resources, diverse sensor networks, and the need for real-time data
processing. It also proposes future directions, such as developing lightweight, energy-efficient algorithms
that operate effectively within the constrained environments typical of IoWT applications. Integrating
ML and DL methods strengthens IoWT security while protecting and preserving wetlands through
intelligent and resilient systems. These findings offer researchers and practitioners valuable insights into
the current state of IoWT security, helping them drive and shape future advancements in the field.

This study analyzes climate patterns in the Kurgan and Chelyabinsk regions of Russia using highresolution data from the Climate Research Unit (CRU) between 1990 and 2020. The research focuses on
how temperature and precipitation have evolved over time and their impacts on local ecosystems,
agriculture, and water resources. Using MATLAB for visualization, temperature and precipitation maps
were created for January and July across five time periods to understand the spatial and temporal
variations in these regions. The analysis revealed a noticeable increase in temperature, with warmer
winters and hotter summers in both regions. Precipitation patterns showed a shift, with a significant
decrease in winter precipitation (January) and an increase in summer rainfall (July). These findings
highlight the need for targeted regional climate adaptation strategies to cope with changing weather
patterns and ensure sustainable management of resources.

The facial morphing method combines at least two images of the face to get a singular altered facial
image that exposes the vulnerabilities of face recognition systems (FRS). The extensive implementation
of face recognition algorithms, particularly in Automatic Border Control (ABC) systems, has raised
apprehensions over potential threats, as modified passports present significant risks to national security.
In this paper, a new face morphing attack detection approach has been proposed using two different
datasets (StyleGAN and AMSL) for testing and validation. A new model for face morphing attack
detection based on a special Convolutional Neural Networks (CNNs) architecture has been used for
classifying real and morphing images. Experimental results indicate that our suggested approach is
extremely generalized and markedly robust in detecting face morphing attacks produced by different
technique

In this paper, geometric methods and wave optics were used to calculate the basic profiles and
characteristics of the lens antennas. The planar reflector arrays were assembled using an iterative
method with multiple forward and inverse Fourier transform calculations. Three-dimensional
electromagnetic modeling was performed in CST Microwave Studio software to evaluate the technical
parameters of the designed antennas. Measurements of the characteristics of fabricated prototypes of
far-field scanning antennas were performed using a custom-designed experimental setup. This study
focuses on the analysis and development of scanning antennas for use in millimeter-wave wireless
communication systems. The researchers aim to develop a lenticular antenna in the 57.24-65.88 GHz
band with a high DER and a wide-angle radial scanning function in the azimuth plane and another
lenticular antenna system in the 58-62 GHz band with a high DER and a wide scanning sector in the
azimuth plane. The researchers also aim to develop a methodology to design flat reflective gratings to
provide electron beam scanning in the azimuthal plane. The characteristics of prototypes of these
antennas have been studied, and experimental tests have been conducted. The study predicts the
realization of scanning antennas in the 57.24-63.72 GHz band using microstrip flat reflector grids.

Data-driven decision-making, real-time connectivity, and automation have transformed industrial
operations with the Industrial Internet of Things. However, the integration also introduces substantial
cybersecurity vulnerabilities, making IIoT networks a prime target for malicious activities. Cyber threats
are evolving and becoming more sophisticated, which makes traditional security mechanisms
inadequate. An approach using deep learning to detect malicious activities in IIoT environments is
examined. It is investigated whether Deep Feed Forward neural networks, autoencoders, and
convolutional neural networks are effective at detecting anomalies and mitigating cyber threats. NSLKDD and UNSW-NB15 benchmark datasets are used to evaluate the proposed model's accuracy,
precision, and detection rates. In addition to strengthening IIoT security, deep learning techniques can
also ensure the resilience of industrial infrastructure.

The rapid digital transformation of education, driven by the widespread adoption of smart devices and
online platforms, has ushered in the era of smart education. While this shift enhances learning
experiences, it also introduces significant cybersecurity risks that threaten the confidentiality, integrity,
and availability of educational resources, student data, and institutional systems. This survey examines
how deep learning (DL) and computer vision (CV) techniques can enhance cybersecurity in smart
education environments. By reviewing 202 peer-reviewed research papers published between January
2022 and June 2025 across leading publishers such as ACM Digital Library, Frontiers, Wiley Online
Library, IGI Global, Nature, Springer, ScienceDirect, MDPI, IEEE Xplore Digital Library, Taylor &
Francis, Sage, BMC, and Google Scholar, the study explores the integration of these advanced
technologies to address emerging threats. It highlights the use of DL in intrusion detection, anomaly
detection, and biometric authentication to protect digital learning platforms. It also examines how CV
techniques, such as facial recognition, behavioral analysis, and emotion detection, enhance security and
foster adaptive learning environments. The survey also addresses key challenges, including data quality,
model interpretability, computational costs, and ethical considerations. By identifying research gaps and
proposing future directions, this survey offers valuable insights for researchers, educators, and
policymakers aiming to develop robust, scalable, and ethical AI-driven cybersecurity solutions for smart
education.

The Adaptive Neuro-Inspired Learning Algorithm (ANILA) offers a breakthrough in the realm of
machine learning by drawing inspiration from the biological processes of the human brain. Developed
to address limitations in conventional models such as CNNs and RNNs, ANILA enhances real-time
responsiveness, energy efficiency, and system adaptability. By emulating neurobiological behaviors
particularly sparse coding and synaptic plasticity ANILA allows systems to process data dynamically,
adjust to novel inputs without retraining, and scale effectively across environments like IoT and
healthcare diagnostics. Performance evaluations highlight significant reductions in latency, increases in
energy efficiency (up to 92%), and exceptional adaptability to changing data streams. Despite current
constraints linked to neuromorphic hardware and the interpretability of its learning processes, ANILA
sets a robust foundation for the development of responsive, scalable, and sustainable AI systems across
diverse sectors, particularly in real-time healthcare monitoring and diagnostic

Since its emergence in late 2019, COVID-19 (Coronavirus Disease 2019) has become one of the most
critical global health threats, claiming millions of lives and placing many more at serious risk. The
complexity of diagnosing COVID-19 lies in the wide range of clinical and examination features involved,
prompting researchers to explore various advanced diagnostic methods. However, one of the main
challenges is identifying the most relevant features that can streamline and improve diagnostic accuracy.
In this study, we propose a feature selection approach based on the Whale Optimization Algorithm
(WOA) to identify key examination indicators associated with COVID-19. We used a dataset of 78
patients that included 25 features, covering demographics, symptoms, vital signs, laboratory findings,
and chronic health conditions. The WOA was applied as a single-objective optimization technique to
select the most informative features. These selected features were then used with the K-Nearest
Neighbors (KNN) algorithm to classify patients into three categories of severity: mild, moderate, and
severe. To evaluate the effectiveness of WOA, we compared it against six other well-established
metaheuristic algorithms: Particle Swarm Optimization (PSO), Multi-Verse Optimizer (MVO), Grey
Wolf Optimizer (GWO), Moth-Flame Optimization (MFO), Firefly Algorithm (FFA), and the BAT
algorithm. Results showed that WOA successfully reduced the feature set from 25 to just 6 key features
while achieving a high classification accuracy of 92.5%. It also demonstrated strong robustness, as
reflected in its low standard deviation compared to other methods. Overall, the proposed WOA-COVID19 framework proved to be a highly effective and efficient solution for feature selection in the context of
COVID-19 diagnosis

Urban traffic control has become increasingly complex with rising vehicle density, particularly in smart
cities. Timely arrival of emergency vehicles is critical, yet existing systems relying on manual
transmitters and sirens offer limited range and effectiveness. This paper proposes a real-time intelligent
traffic management framework integrating Software-Defined Networking (SDN), Internet of Things
(IoT) technologies, and the Edge of Things (EoT)—a paradigm combining edge computing with IoT to
enable low-latency processing at the network edge. The framework connects the SUMO traffic simulator
and Veins vehicular network framework via TraCI, with the RYU SDN controller dynamically adjusting
traffic signals and vehicle routes for emergency vehicle prioritization. Simulation results show that the
system reduces emergency vehicle travel delay by 38%, decreases intersection waiting time by 42%, and
improves overall traffic throughput by 25% compared to conventional control. These results demonstrate
the framework’s scalability, responsiveness, and potential for deployment in real-world smart city
infrastructures

To enhance image dehazing and visual recognition in real-world conditions, we introduce HAZEIMAGE-DATASET, a large-scale dataset comprising nearly 42,000 images. It is constructed from 1,532
clean images sourced globally and captured using a Samsung smartphone, covering diverse natural and
urban scenes. The dataset includes synthetic and real haze variations. Synthetic haze was generated using
MATLAB-based atmospheric scattering models with depth maps for 10 fog levels. Colored haze was
created using alpha blending (α = 0.4) in six colors: red, green, blue, yellow, white, and black. Low-light
conditions were simulated via uniform darkening at 10 levels. Also, 616 real haze images were captured
using a steam device to replicate genuine haze characteristics. Benchmarking was performed using
MobileNetV2 and GoogLeNet. On a hazy subset, MobileNetV2 improved PSNR from 13.34 dB to 21.77
dB and SSIM from 0.805 to 0.950, Also, GoogleNet achieved a PSNR of 22.38 dB and SSIM of
0.953.Classification experiments using a seven-class subset showed high accuracy: 98.93% with
AlexNet and 98.13% with GoogLeNet.The dataset is organized into haze types, levels, and color folders.
It serves as a strong foundation for training and evaluating image dehazing and recognition models under
different visibility conditions. It is available on GitHub

In 2015, the United Nations adopted the Sustainable Development Goals (SDGs) to end poverty, protect
the planet, and ensure global peace and prosperity by 2030. However, progress has been hindered by
challenges like the COVID-19 pandemic, climate change, funding shortages, political instability, and
data limitations. Municipal services, crucial to achieving the SDGs, provide essential functions like waste
management, healthcare, and public safety. Artificial Intelligence (AI) offers innovative solutions to
enhance these services, improving urban sustainability and fostering public-private collaboration. This
research examines AI's role in municipal services, analyzing its applications, benefits, challenges, and
future potential through case studies and expert insights

The paper fills in a gap in the literature that demonstrates an insufficient number of sturdy detection
schemes that can recognize the small semantic aberrations inherent in LLM-generated deceptive text.
Our proposed co-design hybrid model is Semantic Anomaly Detection with Isolation Forest (SADI)
model that combines the synergistic mixture of a fine-tuned transformer-based LLM for deep semantic
feature extraction with Isolation Forest algorithm that detects anomalies efficiently. This study
introduces SADI, an adaptive semantic-anomaly detector for large-language-model phishing emails.
Using a corpus of 10 000 messages, SADI attains an F1 score of 0.981 (95 % CI 0.978–0.984) and
processes a single message in 18 ms on consumer GPUs. An expanded evaluation against three public
benchmarks and a live enterprise feed confirms robustness to prompt variation. Code, data splits, and a
reproducible environment file accompany the paper. We also prepared the new, more challenging target
dataset, that is phishing attacks synthesized by a variety of state-of-the-art LLMs, denoted LLM-PhishSynth-2025, with this objective in mind. The results of our experiments on three publicly available data
sets and our new corpus show that SADI received a higher F1- score of 0.981 compared to the baseline
models, including separate fine-tuned LLMs, by a wide margin. The proposed SADI architecture is the
first to combine semantic anomaly detection with adaptive, contamination-aware isolation in the context
of LLM-generated phishing, addressing both scalability and evolving attack sophistication. The
theoretical impact is the new architecture of architectural fusion that branched semantic anomaly
detection and the practical advantage of a more robust defense solution against an ever-evolving threat
in cyber space in addition to the provision of a new benchmark dataset to the research community. This
approach is an efficient and scalable solution to combat the wave of the phishing campaigns generated
by AI

The rapid proliferation of social media platforms has greatly amplified the dissemination of fake news,
representing significant obstacles to public trust and evidence-based decision-making, particularly for
the Arabic-speaking population. Meeting the challenge of Arabic fake news detection is a problem
compounded by the complex morphological nature of the language, as well as limited resources. This
study presents a hybrid deep learning framework that integrates two Bidirectional Gated Recurrent Units
(BiGRUs) along with an attention mechanism for efficiently detecting misinformation in Arabic news.
The method leverages FastText word embeddings for disambiguating the intricate semantics of the
Arabic language. The model is meticulously crafted to account for the morphological variability and
contextual sensitivities of Arabic, with the extensive Arabic fake news dataset (AFND) being used for
training and testing. Experimental findings show that our model performs better with an accuracy of
91.92%, outperforming current state-of-the-art approaches. The findings highlight the effectiveness of
integrating advanced neural architectures and tailored preprocessing for Arabic, paving the way for more
robust and interpretable fake news detection systems in low-resource languages.

Automatic video colorization remains a challenging computer vision task, particularly when ensuring
semantic accuracy and temporal coherence across dynamic, multi-scene content. Existing methods
often rely on a single fixed reference image, which fails to adapt to abrupt scene changes or variations
in lighting and texture. This study presents a hybrid deep learning framework that dynamically selects
multiple reference images per scene using adaptive thresholds derived from the Structural Similarity
Index Measure (SSIM) and deep features extracted via a ResNet50 backbone with Generalized Mean
Pooling (GeM). The framework integrates three specialized modules pre-processing, reference image
processing, and attention-based colorization—operating in the Lab color space before conversion to
RGB. Experimental evaluations on the YouTube-8M dataset demonstrate a PSNR of 37.89, SSIM of
0.998, and inference speed of 2.6 FPS with a compact 81 MB model (3.2M parameters). Compared
to state-of-the-art methods, the proposed approach achieves superior color fidelity and temporal
stability while maintaining efficiency, making it suitable for deployment in resource-constrained
environments such as embedded vision and IoT systems

Diabetes is a widespread disease worldwide that does not differentiate between children and adults. It
also affects the elderly and pregnant women. However, early detection of the disease facilitates its control
to avoid the effects resulting from delayed diagnosis. With the emergence of artificial intelligence
represented by machine learning techniques and its use in most sectors, accordingly, the adoption of
machine learning techniques to help in disease prediction has become a necessity. This study proposes a
machine learning algorithm-based approach for diabetes prediction. This study uses three datasets, two
of which are private and the other includes the Pima Indians dataset. Six machine-learning models have
been used and evaluated in this study including Gaussian Naive Bayes model. Bernoulli Naive Bayes
Model, Bagging Regressor Model, Neural Network Architecture, Multilayer perceptron, and SVR. To
address the imbalance in classes of the private datasets, the SMOTE technique has been utilized. To
analyze the state of the arts, a systematic literature review was conducted. The results showed that the
Bagging Regressor algorithm is the best among the used algorithms in terms of the accuracy of the
derived results. It achieved an accuracy of 99.79 with SMOTE included and 97.95 without SMOTE. A
smart mobile application was developed based on the proposed approach that facilitates clinicians to
predict diabetes. This study strengthens the theoretical foundations of machine learning in healthcare by
presenting a robust and empirically validated approach for early detection and prediction of diabetes. The
findings not only advance academic knowledge but also provide practical guidance for developing AIbased diagnostic tools in clinical settings.

Transportation networks impact millions of people daily. Their efficiency immediately affects travel
time, safety, and environmental sustainability. Unfortunately, various issues hinder the expected
performance and efficiency of these networks. Traffic congestion is an up-to-date issue in the urban
environment. Fuel consumption is high because travel time has increased, which has a passive
environmental impact. Extensive research has been conducted to progress the intelligent transportation
systems installed on communication networks and information to treat this congestion. However, there
is a significant amount of affront residue in combining real-time data, estimation analytics, and 5G
abilities effectively. This paper offers a novel routing algorithm integrating vehicular ad hoc networks
with 5G technology to increase routing efficiency and minimize congestion. This routing is named 5G
adaptive traffic management (5G-ATM). It collects real-time data from connected vehicles and roadside
units to estimate traffic status and congestion. Out of simulations in an urban environment, the proposed
5G-ATM routing significantly progresses over previous routing protocols, such as an ant colony-inspired
energy-efficient for optimized link state (AC-OLSR) routing and directional-cache agent-based locationaided (D-CALAR) routing. During rush hours, 5G-ATM shows the lowest traffic congestion events.
Moreover, it minimizes average travel times by almost 8% compared to D-CALAR and 21% compared
to AC-OLSR. These outcomes suggest that combining vehicular ad hoc networks with 5G technology
helps manage traffic more efficiently, providing an efficient pathway and practical transportation
systems.

One of the most important fruit crops in the world is citrus. However, some citrus diseases spread rapidly,
which is why early detection at an accurate stage is important for timely intervention. YOLO-based
object detection models, such as the latest YOLOv10, where small lesions are difficult to identify among
noisy backgrounds, have recently been developed, yet their accuracy tends to degrade. Therefore, we
proposed a citrus disease detection model by integrating the region of interest (ROI) for object
segmentation with the YOLOv10 model, thus addressing the issues of low detection accuracy and slow
inference time. The proposed model was trained on an annotated dataset of three major citrus pathologies:
anthracnose, citrus canker, and leaf miner infestation. Results showed significant improvements in the
performance of our proposed model, which incorporates the ROI mechanism into YOLOv10.
Specifically, the ROI-YOLOv10 model achieved high mAP scores, reaching 0.99 and 0.985 during
training and validation, respectively, and maintaining high generalization capabilities with a test mAP of
0.984. Precision and recall metrics similarly underline the enhanced accuracy and robustness of ROIYOLOv10. Compared with previous YOLO-based studies, our model exhibits enhanced accuracy and
faster inference times. The incorporation of ROI techniques into the YOLOv10 framework is a highly
effective approach for improving agricultural productivity by facilitating early and precise detection of
plant disease.

Automatic license plate recognition (ALPR) is essential for intelligent transportation systems, traffic
monitoring, and law enforcement. Although deep learning has significantly progressed ALPR, the
adaptation of these techniques to the specific characteristics of Iraqi plates presents challenges due to the
variety of fonts, differing plate dimensions, and intricate real-world driving conditions. This paper
presents an end-to-end Automatic License Plate Recognition (ALPR) framework specifically designed
for Iraqi license plates. This system integrates a Detection Transformer (DETR) for the identification of
plates alongside a Convolutional Recurrent Neural Network (CRNN)-based Optical Character
Recognition (OCR) module for the purpose of character recognition. The system is trained and evaluated
on a newly developed dataset of 1,000 annotated images representing diverse driving scenarios in Iraq.
It achieves a mean Average Precision (mAP@[.5:.95]) of 0.91 for detection and a full-plate OCR
accuracy of 93%. The accuracy of DETR surpasses that of YOLOv5, Faster R-CNN, and SSD, as
demonstrated by comparative experiments. Conversely, a lightweight transformer (DeiT-Tiny-Det)
approaches DETR's performance at faster inference velocities, illustrating a practical trade-off between
speed and precision. Ablation studies confirm the importance of robust detection for end-to-end accuracy,
while error analysis shows that low-light and character-level confusions remain the main challenges. The
results indicate that transformer-based detectors, in conjunction with specialised OCR models, yield
dependable region-specific ALPR appropriate for real-world application

In digital images, protecting sensitive visual information against unauthorized access is considered a
critical issue; robust encryption methods are the best solution to preserve such information. This paper
introduces a model designed to enhance the performance of the Tiny Encryption Algorithm (TEA) in
encrypting images. Two approaches have been suggested for the image cipher process as a
preprocessing step before applying the Tiny Encryption Algorithm (TEA). The step mentioned earlier
aims to de-correlate and weaken adjacent pixel values as a preparation process before the encryption
process. The first approach suggests an Affine transformation for image encryption at two layers,
utilizing two different key sets for each layer. The dual encryption process achieves high diffusion and
confusion properties for the cipher process. The second approach proposed a chaotic Arnold map before
the Tiny Encryption Algorithm (TEA) process. Various statistical measures are used, such as Peak
Signal to Noise Ratio (PSNR), Structural Similarity Index (SSIM), and Image Quality Index (IQI). For
example, the lower PSNR, SSIM, and IQI values indicate better results for test image Lina of the second
approach. The obtained results for the previous measures of the second approach are 8.5449, 0.0008,
and -0.0061, compared to the first approach, 8.5529, 0.0054, -0.0015, respectively. Moreover, key space
and time analysis are used to assess the encryption process. The outcomes show a high-key space
(32,768 ∗ 2
128) and a slight encryption time of 130 milliseconds for the first approach and 1862
milliseconds for the second approach

The Internet of Things (IoT) has significantly transformed modern systems through extensive
connectivity but has also concurrently introduced considerable cybersecurity risks. Traditional rulebased methods are becoming increasingly insufficient in the face of evolving cyber threats. This study
proposes an enhanced methodology utilizing a hybrid machine-learning framework for IoT cyberattack detection. The framework integrates a Grey Wolf Optimizer (GWO) for optimal feature
selection, a customized synthetic minority oversampling technique (SMOTE) for data balancing, and a
systematic approach to hyperparameter tuning of ensemble algorithms: Random Forest (RF), XGBoost,
and CatBoost. Evaluations on the RT-IoT2022 dataset demonstrate that GWO reduces features from
32 to 21, thereby enhancing computational efficiency and interpretability without compromising
accuracy, while customized SMOTE addresses class imbalance and enhances minority-class detection.
The optimized RF and XGBoost models were assessed using accuracy, precision, recall, and F1-score
metrics, and achieved 100% accuracy with strong generalization. These results highlight the
effectiveness of optimization-based feature selection and data balancing in improving IoT security that
is extensible to deep learning and ensemble-based approaches.

The increase in cyber-attacks makes the picture protection problem more urgent, insisting on finding
new defensive means and remedies. This paper presents such a remedy, called DGEN, short for
Dynamic Generative Encryption Network. This mechanism could very well signify a substantial leap
forward in picture protection. To understand what DGEN is doing and why, we must take a slightly
longer journey to comprehend the basic idea it uses to attain the protection it claims to offer. DGEN is
to a set of pictures what a good secure lock is to your front door. It could be better than a strong
encryption algorithm, like AES, that serves as a static lock. Indeed, studies run by the authors of this
paper indicate that DGEN serves as a mechanism that could encrypt the set of images with much better
performance in terms of security, flexibility, and maintainable low computational cost.

A lightweight digital twin model for a single 6G cell operating in the D-band (140 GHz) with a 1 GHz
bandwidth is presented in this work with the goal of assessing the cell's capacity, coverage, and terminal
time in order to support extended reality (XR) applications. With a tangent dispersion of 3 dB and a path
exponent of n = 2 .2, the model is based on the free-space loss equation as per ITU-R Recommendation
P.52 5. The instantaneous capacity is determined using the Shannon-Hartley theorem. Three XR sessions
are created every minute using a Poisson method, and their durations are determined by an exponential
distribution (mean of 12 0 seconds). In accordance with 3 GPP and Ericsson guidelines for normal XR
loads, the bit needs per user are randomly selected to fall between 40 and 12 0 Mb/s. The average coverage
was around 92 %, the average cell capacity was approximately 5.1 Gb/s, and the edge capacity (lowest
quintile) was approximately 230 Mb/s, according to fifty statistical forecasts. Additionally, the 95th
percentile round-trip latency was 3 .9 ms, which is significantly less than the permitted maximum (10–
20 ms) for immersive XR research. These findings suggest that modest XR loads may be supported by a
250-meter cell with a high-gain antenna layout without the need to immediately lower the radius or raise
the transmitted power. However, the model remains theoretical and simplified, excluding geometric
blockage and cell overlap in complex metropolitan environments

Immersive learning using Virtual Reality (VR) has gained prominence for delivering experiential, engaging
education. However, most VR learning environments lack real-time adaptability, personalization, and cognitive
responsiveness. This study presents an Agentic AI-enabled VR framework that autonomously adjusts pedagogical
content, interaction style, and challenge level based on learner behavior, emotions, and performance feedback. The
proposed system integrates a reinforcement learning-based agent with a virtual reality module to form an intelligent
tutor capable of independent decision-making. A neuro-symbolic model processes multi-modal feedback (gesture,
speech, gaze, performance) to determine context-aware pedagogical strategies. The system employs a self-evolving
curriculum logic that adapts in real time. Experiments are conducted using Unity3D integrated with Python-based
RL agents and simulated student models. Results demonstrate a 35.7% improvement in learner retention and a
42.1% reduction in cognitive overload compared to traditional static VR systems. The agent successfully
personalized 92.3% of scenarios without human intervention. Emotional adaptivity and dynamic pacing showed
increased engagement and reduced frustration metrics among diverse learners. Agentic VR represents a paradigm
shift in intelligent education systems, enabling autonomous, emotionally-aware, and responsive learning
environments. The proposed framework outperforms conventional VR platforms by offering real-time adaptive
learning without pre-scripted logic. Integrating reinforcement learning, neuro-symbolic reasoning, and affective
feedback into a single VR space results in a novel contribution to adaptive educational technology. The reported
metrics were obtained from an administrator-controlled user study of 10 participants (aged 18 - 25 years), all of
whom participated in three VR-based learning experiences as described in a controlled evaluative protocol. This
study provides a foundation for future work on autonomous agents in educational met verses, special education,
and lifelong learning systems.

Due to the widespread popularity of digital images on the Internet, image-based steganography has
become a widely adopted technique for embedding secret information into everyday visual content. In
parallel, steganalysis plays a vital role in digital forensics and information security by seeking to uncover
hidden content within these images. Although steganographic techniques—particularly those employing
adaptive embedding strategies—have made significant progress, many steganalysis approaches still
struggle to generalize effectively across different image types and embedding methods. This contrast
highlights the need for more intelligent, flexible, and robust analysis frameworks. This review examines
steganographic techniques for digital images and the application of metaheuristic algorithms in
steganalysis. These algorithms are employed in tasks such as feature selection and parameter
optimization. They can also function as classifiers themselves, thereby enhancing the detection of hidden
information. We conducted a structured review of over 100 research articles, categorizing steganographic
approaches based on their embedding domain (spatial and frequency) and steganalysis techniques
according to the metaheuristic algorithms they utilize. Metaheuristic algorithms have demonstrated
significant promise in improving the effectiveness of steganalysis by optimizing both feature selection
and classification processes. However, their performance is often influenced by factors such as parameter
tuning, initialization strategies, and the quality of extracted features. Recent studies also show a growing
trend toward hybrid and ensemble-based techniques, which further enhance detection accuracy and
reliability