Iraqi Journal for Computer Science and Mathematics

In this paper, a new lightweight U-Net deep learning-based neural network designed for the segmentation of skin
lesions is proposed. Segmentation of skin lesions is the most critical step in computer-aided dermatology diagnosis for
the early detection of melanoma and other diseases. However, we address the difficulty related to the precise definition
of the lesion margins with an eye on the computation cost. We have demonstrated the state-of-the-art performance of
DeepSkinSeg in most metrics on dermoscopic images using the PH2 and Human Against Machine (HAM10000) datasets.
The metrics of the DeepSkinSeg model were robustness measured as the Intersection over Union (IoU) at 91.49, Dice
coefficient at 95.56, precision at 97.97, sensitivity at 96.84, and accuracy at 96.71 for the PH2 dataset. Other standard
generalization capabilities for the HAM10000 dataset could be an IoU of 92.97, a Dice coefficient of 96.36, precision
at 97.64, sensitivity at 95.10, and an accuracy of 94.59. DeepSkinSeg has a very efficient inference because the model
itself is lightweight, proving to be very helpful for real-time dermatological analysis. This work further advanced the
computer-aided diagnosis in the task of skin lesion classification, guaranteeing even more promising clinical applications.

This study employs the Young Hussein Jassim (YHJ) technique to examine the exact solutions of the space-time
telegraph equation analytically (ST-TE). The YHJ approach is an innovative and appealing hybrid transformation
integration method, effectively combining the HJ and Young methods. Through a simplified iterative process with
minimal computational requirements, this approach quickly provides convergent, sequential solutions. The reliability
of the method is demonstrated by applying it to two case studies of the ST-TE within the framework of the Tania
derivative, which includes the definition of non-singular kernel functions. The study also includes extensive comparisons
between approximate, exact, and relevant literature-based solutions to assess the technique's accuracy and effectiveness.
Graphical representations illustrate the effect of non-integer temporal and spatial parameters on solution behavior. The
findings suggest that this method is easy to implement and suitable for exploring complex physical models governed by
nonlinear partial differential equations with fractional time components.

Accurate classification of cardiovascular diseases (CVDs) is of utmost importance for cardiologists to provide ap-
propriate treatments. Diagnosing and predicting cardiovascular conditions are crucial medical responsibilities in this
context. The healthcare sector is increasingly utilizing deep learning (DL) and machine learning (ML) algorithms due
to their ability to identify patterns in data. Diagnosticians may reduce the number of misdiagnoses by using DL and
ML techniques for the categorization of cardiovascular disease incidence. To reduce the mortality linked to CVDs, this
research offers a unique model that properly predicts and classifies these problems. This research presents approaches
such as deep learning, random forest (RF), support vector machines (SVM), and K-nearest neighbors for predicting heart
disease arrest. We implemented the suggested model using 303 real-world instances from Kaggle. In the testing stage,
the KNN model’s accuracy was 92%. By comparison, the accuracy of the DL model was 87%. The RF model’s accuracy
was 84%. The results indicate that the KNN algorithm outperforms other algorithms in terms of accuracy. We compared
the study’s results with a variety of existing systems.

Texture features and stability have generated significant interest in biometric recognition. The inner knuckle print
is distinctive and difficult to fake, making it extensively used in individual identification, criminal investigation,
and various other domains. In recent years, the rapid progress of deep learning technology has created new prospects for
internal knuckle recognition. This paper proposes a robust inner-knuckle-print recognition system (RIKP-RS) depending
on two deep learning (DL) models. This paper focuses on the key components of the inner surface of the hand namely the
little finger, ring finger, middle finger, index finger, and thumb finger that are used for human identification. Using the
new segmentation method, rely on the Hands Landmark Module (MediaPipe Module) to detect components that have
important biometric features. By Considering the inner knuckle print (IKP) as a texture, this study adopts two effective
models: The DenseNet201 model and the InceptionV3 model to extract distinctive features from every modality. Uses all
the key points of inner knuckle prints (IKP) of ten fingers for concatenated fusion recognition of all the features extracted
by these models. Ultimately, these features are classified by different similarity metrics that are employed to compute
the matching procedure for each model individually. A dataset of 11,076K hands with left and right palms was used to
evaluate the proposed system. The system achieved the best performance on this dataset with a rank-1 score of 98.45%
on the denseNet201 model, a rank-1 score of 99.81% on the inceptionV3 model for all left IKP, a rank-1 score of 96.68%
on the denseNet201 model, and rank-1 score of 98.32% on the inceptionV3 model for all right IKP. These results cover
the inceptionV3 model for all concatenated fusion recognition. In terms of performance, the RIKP-RS outperforms the
most advanced inner knuckles pattern (IKP) recognition systems.

Count data modeling requires usage of the Poisson regression model as a primary analytic method. Excess dispersion in variables makes the model unfit to use when the Poisson distribution mean value differs from its variance value. Data f its well with the results obtained by using the Bell regression model. Excess zeros occur frequently in the observed count data records. The Zero-Inflated Bell regression model is a substitute for the Bell regression model in this situation. The approach of maximum likelihood is mostly used to estimate the Zero-Inflated Bell regression model’s parameters. When modeling the link between the response variable and two or more explanatory variables in an extended linear model, such as the Zero-Inflated Bell regression model, linear dependency poses a risk in a real-life application. It decreased the greatest likelihood estimator’s effectiveness. To address this problem, we proposed a new ridge estimator for the Zero-Inflated Bell regression model. The results of the simulations and implementations validate the suggested approaches’ superiority to the traditional maximum likelihood estimator.

Arabic script is exhibited in a cursive style, which is a departure from the norm in many common languages, and
the shapes of letters are contingent on their positions within words. The form of the first letter is influenced by the
subsequent letter, middle letters are shaped by both preceding and succeeding letters, and the shape of the final letter
is determined by the preceding letter. Additionally, certain letters are found to have strikingly similar shapes, making
Arabic text recognition a formidable challenge in computer vision. The challenge of detecting and recognizing Arabic
handwritten text is addressed in this paper by proposing a novel system that integrates two Deep Learning models
without the constraints of a predefined dictionary or language model. Object Detection techniques are employed in the
first model to accurately identify lines, words, and punctuation marks, providing a robust foundation for text recognition.
In the second model, a powerful recognition system is developed, trained on the IFN/ENIT dataset to predict character
sequences from handwritten text images. The architecture is comprised of a convolutional neural network (CNN) with
residual connections, followed by a Bi-directional Long Short-Term Memory (BLSTM) layer, and culminates in a fully
connected layer. State-of-the-art results in unconstrained Arabic text recognition tasks are achieved by this approach.

Finger vein authentication is a biometric technique that uses unique vein patterns within the finger to verify
identity, ensuring security by scanning veins with near-infrared light to prevent counterfeiting. The reliability of this
technology and its resistance to skin disorders make it highly appreciated. The motivation for using PCANet deep
learning in finger vein authentication arises from the need for a more precise, secure, and effective biometric system.
Conventional methods have trouble with image quality, noise, and illumination, whereas PCANet improves classification
accuracy by employing principal component analysis (PCA) to extract deep features. Its lightweight structure guarantees
computational efficiency while maintaining a high level of security, making it ideal for applications that require robust
authentication. By using high-quality data and lowering the computational complexity needed by the PCANet approach,
the suggested system seeks to increase performance. This research introduces three approaches: high-frequency emphasis
filtering (HFEF), gabor filter (GF) images, and novel enhanced image techniques using contrast-limited adaptive
histogram equalization (CLAHE) to keep edges and increase image quality. Second, for dimensionality reduction and
feature extraction, a novel integrated PCANet as well as feature-level principal component analysis (PCA) is suggested.
Ultimately, a pair of authentication models are chosen: one employing hyperparameter machine learning (HPML) in
conjunction with Multi-Layer Perceptron (MLP), K-Nearest Neighbors (KNN), and Support Vector Machine (SVM) for
random search, and the other using the same techniques for simple search. PLUSVein-Contactless finger vein dataset
was used for evaluating such techniques. The suggested approaches performed better on this dataset, with an F1 score
of some metrics of 100% for SVM, 99.90% for KNN, and 100% for MLP. Those results were for the random search
model with PCA. The outcomes demonstrated a considerable improvement in F1-score of finger vein authentication
using PCANet DL, both without and with PCA. Furthermore, the suggested authentication mechanism outperformed
previously suggested approaches, in terms of improved image quality, good vein-specific features, and obtaining the
best set of optimal parameters for matching.

When ensuring the reliability of device or the suitability of a material, it is necessary to take into consideration the
stress cases in the operating environment. This means that the uncertainty about the reality environmental stress must be
taken into as random. The stress-strength (S-S) model treated the stress and strength variables as random. In the simplest
form of stress-strength model, y represents the stress put on the unit by the operating environment, and the strength of
the unit represented by x. A unit is able to perform its required function if its stress imposed on it is less than the strength
of the unit. In this paper, the stress-strength reliability estimation for the modified exponentiated Lomax distribution,
which is generalization of the Lomax distribution, with an unknown shape parameter and a known scale parameters is
studied using different methods. These methods include the maximum likelihood method, Bayesian estimation method
under a quadratic loss function, and the least squares method for complete data. The estimators are compared based on
Markov Chain Monte Carlo (MCMC) simulations using R-Studio, evaluated by the mean square error (MSE) criteria. The
simulation results show that the maximum likelihood estimators are the best in two cases: the first is when the sample
sizes are equal and the second is when the shape parameter of the strength variable is greater than the shape parameter
of the stress variable. While least squares estimators are the beast if the strength sample size is smaller than the stress
sample size. Finally when the strength sample size is greater than the stress sample size, then the best estimators differ
between the maximum likelihood estimators and Bayesian estimators. Bayesian estimators become the best when the
shape parameter of stress variable is larger than the shape parameter of the strength variable.

Quantum Learning (QL) has emerged as a promising approach to medical image classification, leveraging the principles
of quantum mechanics to improve the performance and efficiency of machine learning algorithms. This systematic
review provides a comprehensive critical assessment of the current status of QL techniques developed for medical image
classification, with a specific focus on trends, methodologies, and future directions in this rapidly evolving field. A
thorough literature search was conducted across five major databases, resulting in a total of 28 relevant studies published
between 2018 and 2024. The studies were analyzed and classified based on the type of quantum algorithm, the medical
image modality, and the performance metrics used. The analysis revealed a diverse range of QL techniques, including
Quantum Support Vector Machines (QSVM), Quantum Convolutional Neural Networks (QCNN), and various hybrid
quantum-classical approaches. These techniques have been applied to diverse medical image classification tasks, such as
brain tumor classification, skin lesion classification, and COVID-19 detection, demonstrating promising results in terms
of accuracy, sensitivity, and specificity. However, several challenges were identified, including the preprocessing and
encoding of medical images for quantum processing, the limited scalability of current quantum hardware, and the need
for interpretable and explainable QL models. This review underscores the immense potential of QL to revolutionize
medical image classification, while also emphasizing the necessity of multidisciplinary collaborations and further
research to overcome existing challenges and facilitate the integration of QL techniques into clinical practice.

In this paper, we discuss the numerical solution of fractional order partial integro-differential equations. The type
of fractional derivative used is a Caputo derivative. The method proposed in this paper known as transform optimal
perturbation iteration method. This method combines the optimal perturbation iteration method and the Laplace
transform in order to converge to the exact solution. The proposed method is highly efficient and provides the means
of controlling the approximate solutions convergence. Illustrative examples prove that the suggested approach is very
accurate when compared with the exact solution and the results of the existing methods.

Evaluating company growth potential has moved away from traditional financial focused ratios and ratios analysis
that has origins in the early twentieth-century economics. However, these conventional methods might not be accurate
in measuring such efficient factors as this combined proposed framework of Data Envelopment Analysis (DEA) and
improved mathematical models do. The present research focuses on the prospect of growth of companies through
evaluating the performance of 40 DMUs in terms of efficiency DEA and MMTs. DEA is used to determine the efficient
DMUs while SFA underline the factors such as investment on research and development, effective marketing strategies
and qualified human resource as the determinants of efficiency. Bootstrap Confidence Intervals (BCIs) are employed for
the purpose of improving the efficiency scores precision. The integration of DEA with MMTs presents useful information
on the identification of the key drivers of the growth and the effective formulation of efficient and competitive strategies
for the managers and policymaker. The study emphasizes the need to apply these combined methodologies as valid in
the growth evaluation of companies with implications to strategic decisions. This proposed framework can be used in
finance, marketing, healthcare, and operations management to compare the current organizational effectiveness and
future development.

The Software-Defined Networking (SDN) paradigm decouples the control and the data plane. One of the most
significant challenges in this paradigm is SDN controller placement optimization, since improper placement may
dramatically influence latency, load balancing, and network resilience. The paper proposes the Improved Harmony
Search Algorithm (IHSA) as a new approach for the placement of SDN controllers. To overcome the disadvantages
of conventional optimization methods like HSA, GA, and PSO, adaptive parameter tuning, dynamic harmony memory
management, and updating rules for enhanced memory are incorporated into the IHSA. The complete simulations of IHSA
over small, medium, and large-scale SDN topologies have shown its robust supremacy in performance compared to the
benchmark algorithms, which provide up to 23.5% less latency along with significant improvements in load-balancing
and convergence-time. The optimization results of IHSA are well supported by the statistical validity based on the
p-values and confidence intervals. The algorithm is scalable, adaptive, and very efficient. Hence, it makes a more realistic
solution for real-time and dynamic SDN environments. The results show that IHSA has the potential to improve network
performance in latency-sensitive applications such as IoT, cloud computing, and telecommunications, thus opening up
resilient and high-performance network infrastructures. This is a practical application of IHSA, and subsequently, the
recommendation given through this study is for its integration with datasets and machine learning techniques from real
world to further improve optimization.

Deep learning’s rapid development is generating significant interest in its potential to improve medical imaging. It
has shown promising results in detecting malignant lymphoma in histopathology medical images. Image classification
methods are widely used to aid in making diagnoses from medical images. In recent years, deep learning methods
have achieved high performance in detecting malignant lymphoma in histopathology images. This study proposes
a novel approach to improving lymphoma diagnosis in histopathology images called the Lightweight Convolutional
Neural Network (LWCNN). The proposed LWCNN model comprises multiple deep learning architectures, including a
convolutional neural network (CNN) that has been trained to classify lymphoma subtypes based on histopathology
images using ResNet50 and MobileNetV2. The LWCNN model aggregates the predictions of disparate architectures
to arrive at a definitive diagnosis, leveraging the unique capabilities of each. A comprehensive dataset of annotated
lymphoma histopathology images was assembled for the purpose of training the multi-deep learning model. To ensure
a representative and diverse training set, the dataset was meticulously curated to encompass various subtypes of
lymphoma. Performance evaluation of the proposed deep learning model for lymphoma classification using standard
metrics revealed the following accuracies: LWCNN (97.34% training, 86.71% testing), ResNet50 (88.76% training,
86.37% testing), and MobileNetV2 (88.47% training, 86.60% testing). These results indicate that the LWCNN model
significantly surpasses existing approaches in diagnostic accuracy.

Internet service today greatly affects individuals, companies, and organizations around the world, as the Internet
contributes to many things such as facilitating procedures, reducing effort, and saving time. The instability of the
Internet system is a major obstacle to the successful implementation of institutional plans, leading to many workflow
problems, including delays in providing services to citizens and insufficient communication between the components
of the institution. It also leads to a lack of information needed for decision-making, which negatively affects customer
satisfaction and operational efficiency. There is a gap in the literature regarding the evaluation of the relationships
between Internet service criteria and institutional performance. The Iraqi Martyrs Foundation (IMF) sponsors up to one
million citizens from the families of martyrs, which necessitates working on automating operations to accommodate
these huge numbers. This study aims to investigate the effects of Internet service criteria on the performance of
the Iraqi Martyrs Foundation. The study included developing a survey questionnaire to collect data from relevant
stakeholders (technicians, engineers, managers, administrators, buyers, and users) and analyzing it using structural
equation modeling. The results indicated support for the hypotheses that criteria such as quality of service, reliability,
responsiveness, and experience have a direct impact on organizational performance, while the hypotheses that security
and price have an impact on organizational performance were rejected. The resulting model could help improve the
performance of the IMF by defining clear criteria for Internet services.

This paper gives a definition of a new class of T-module and T-submodule called an Endo-Restricted Bounded module (submodule) written shortly by Endo-R.B. module (submodule) and present some different approaches to connect this class of module with other types of modules such as: compressible modules, monoform modules, critically compressible modules, retractable modules, and quasi-Dedekind modules. One of the main purpose of this work is to introduce a few new conditions and reveal some properties and corollaries. This paper considered to be another solution or answer for Zelmanowitz’s problem. In fact, an Endo-R.B. T-module plays an important role to this problem and we say under what condition compressible T-module would be a critically compressible T-module so we present a positive solution depend on an Endo-R.B module. The T-homomorphism of compressible and monoform modules join in a nice way with an endomorphism of a T-module that we need it for the definition of an Endo-R.B module. Moreover, an Endo-R.B module gives us directly three different modules and these modules are bounded module, finitely annihilated module, and retractable module. Finally, polyform and fully polyform modules are involved in this article.

It is important to pay attention to develop our security systems, due to the increase in cyberattacks and the
development of their methods and the work to develop quantum computers capable of penetrating the solution of
complex equations of encryption algorithms. The need to develop protection methods has emerged in line with the
development of hackers to ensure the security and safety of users’ data. In this research, work was done on the modern
Li-Fi technology based on comparing the values with Kim’s model and a distance of more than 13 km was reached in
the fresh air. This technology provides security and privacy inside rooms because it sends signals through light, so the
signal cannot be modified from the outside. However, the threat remains through the transmission medium. Therefore,
a hybrid algorithm was created from chacha20 with a key size of 256 and RSA with a key size of 2048, which guarantees
us a high level of security, making it immune to threats. The chacha20 algorithm provides encryption speed and security,
while the RSA algorithm increases the level of security of the algorithm, as the total time taken to encrypt data from
the sender and decrypt it from the receiver was 0.036 seconds. This ensures high and fast performance for this hybrid
algorithm.

It is important to pay attention to develop our security systems, due to the increase in cyberattacks and the
development of their methods and the work to develop quantum computers capable of penetrating the solution of
complex equations of encryption algorithms. The need to develop protection methods has emerged in line with the
development of hackers to ensure the security and safety of users’ data. In this research, work was done on the modern
Li-Fi technology based on comparing the values with Kim’s model and a distance of more than 13 km was reached in
the fresh air. This technology provides security and privacy inside rooms because it sends signals through light, so the
signal cannot be modified from the outside. However, the threat remains through the transmission medium. Therefore,
a hybrid algorithm was created from chacha20 with a key size of 256 and RSA with a key size of 2048, which guarantees
us a high level of security, making it immune to threats. The chacha20 algorithm provides encryption speed and security,
while the RSA algorithm increases the level of security of the algorithm, as the total time taken to encrypt data from
the sender and decrypt it from the receiver was 0.036 seconds. This ensures high and fast performance for this hybrid
algorithm.