Statistics, Optimization & Information Computing

Preface for the Special Issue of International Conference on Computer Engineering and Artificial Intelligence, I2CEAI24

2026-01-18T04:33:07+08:00

We are delighted to present this Special Issue featuring selected and extended papers from the International Conference on Computer Engineering and Artificial Intelligence (I2CEAI 2024), held in Khouribga, Morocco, on September 27-28, 2024. This conference served as a platform for researchers, academicians, and industry experts to exchange innovative ideas, discuss emerging trends, and explore the latest advancements in computer engineering, artificial intelligence, and related fields.

Discrete Orthogonal Moment Descriptors Applied In Image Classification And Object Recognition Enhanced By Machine Learning Methods

2026-01-18T04:33:08+08:00

Orthogonal moments play a crucial role in image analysis, including applications in image reconstruction and the extraction of consistent, robust features, making them essential for modern computational imaging. This study examines the effectiveness and performance evaluation of orthogonal: Tchebichef moments, Krawtchouk moments, Charlier moments, and Hahn moments for image reconstruction, feature description, and their application in image classification and object recognition using advanced machine learning techniques. We present the construction process of orthogonal image moments. Then the deriving of invariant moments based on Tchebichef, Krawtchouk, Charlier, and Hahn polynomials. These moments are then employed to generate feature vectors specifically tailored for image classification and object recognition tasks. To evaluate feature robustness, we conduct extensive classification experiments on two distinct image databases using K-Nearest Neighbors (K-NN) and Support Vector Machine (SVM) classifiers. The results demonstrate that invariant moments are highly effective at capturing discriminative image features and maintain reliable performance under various noise conditions, including salt-and-pepper noise, underscoring their applicability in real-world scenarios.

Feature Selection via Fuzzy Rough Set Theory for Robust Classification: a Review and Comparative study

2026-01-18T04:33:10+08:00

Despite a variety of powerful classifiers available in machine learning today, most of them struggle with processing large-scale real-world datasets. Usually, these datasets contain irrelevant and redundant information that can negatively affect the model’s performance. To overcome this, feature selection has become a commonly used strategy to improve model performance by reducing dataset size while retaining essential information. Some feature selection techniques tend to require more information than what is provided in the given dataset, making them impractical in some cases. Alternatively, completely data-driven methods may lose critical information, as they can mistake vagueness or imprecision in the dataset for irrelevant or redundant features. Fuzzy-rough set theory offers a robust paradigm for tackling uncertainties, having been utilised across various domains, with feature selection being one of its most prominent applications. This paper presents an extensive review of feature selection methodologies grounded in fuzzy-rough set theory, accompanied by an empirical evaluation of multiple techniques to evaluate their effectiveness.

Optimizing Photovoltaic Performance Prediction Using Machine Learning: Analysing the Impact of Environmental Variables in Marrakesh

2026-01-18T04:33:10+08:00

This study focuses on the optimization of photovoltaic (PV) prediction using machine learning (ML) models by analyzing the impact of environmental variables in Marrakech. The research compares two types of meteorological data from satellites and ground stations to assess their respective contributions to forecast accuracy.
The results show that global solar irradiance (G), air temperature (Ta) and wind speed (Wv) are the most influential parameters on energy production, whatever the data source. However, forecasts based on ground-measured data showed slightly higher accuracy, with an R²=0.98 for measured data versus 0.86 for stalietes data, underlining the importance of localized measurements.
Of the scenarios tested, Scenario 1 (all inputs) achieved the highest accuracy, with an R² of 0.98 and an RMSE of 91.39. Scenarios 2 (without Wv) and 4(without DNI) also delivered acceptable levels of accuracy, albeit slightly lower than Scenario 1. These results highlight the importance of integrating localized weather data to improve the accuracy of PV power generation forecasts.

Comparison of filter techniques for feature selection in high-dimensional data

2026-01-18T04:33:11+08:00

Feature selection constitutes a fundamental challenge within machine learning, which has garnered heightened
attention owing to the proliferation of high-dimensional datasets. Filtering-based feature selection methods hold crucial
importance as they can be seamlessly integrated with any machine learning model and significantly accelerate the runtime
of such algorithms. This study investigates the performance of eight distinct filter methods, examining their efficacy
across seven high-dimensional datasets, the classification accuracy was assessed through the employment of support vector
machines and k-nearest neighbor classifiers, and the Wilcoxon test statistic was applied to confirm the observed results
regarding classification accuracy

DDROP: Encouraging Filter Sparsity in Convolutional Neural Networks via Dynamic Dropout for Prunable Models

2026-01-18T04:33:12+08:00

In this paper, we introduce a novel dynamic dropout (DDROP) based method for inducing sparsity in deep neural networks. Our method works by adaptively dropping neurons or filters during the training phase. Unlike other pruning techniques, DDROP determines the probabilities of dropping neurons based on their importance, with the use of a ranking criteria derived from their activation statistics. Furthermore, we incorporate the l1 regularization to suppress the least important neurons, further enhancing the dynamic pruning process. We evaluate the proposed method on standard datasets such as CIFAR-10, CIFAR-100, and ILSVRC2012 and various network architectures, showing the consistent enhancement in the accuracy of the pruned models compared to other techniques. The results obtained from our evaluations highlight DDROP's promise as a strategy for efficient deep neural networks and its ability to achieve structured sparsity, reducing the complexity of the model while keeping a satisfactory performance.

Existence of optimal controls for semilinear systems with a nonreflexive control space

2026-01-18T04:33:13+08:00

In this paper, we study the existence of optimal controls that minimize a given functional. We consider a class of infinite-dimensional semilinear systems, and a functional that depends on a control function u and the associated solution of the semilinear equation. The functional is minimized over a set of admissible controls, that is a convex subset of a nonreflexive control space. Under appropriate assumptions, we derive sufficient conditions for the existence of optimal controls, for two classes of semilinear systems. Thereby, we provide two examples of partial differential equations to highlight the obtained results.

Algorithm-based optimization of spare parts inventory management

2026-01-18T04:33:13+08:00

This research aims to identify the most effective strategy for determining the ideal quantity of spare parts to order during each period, with the ultimate goal of minimizing management costs. These costs encompass various expenses associated with inventory management. To achieve this objective, we present a mathematical model of single-echelon inventory dynamics using a Markov decision model. Additionally, a method based on genetic algorithms is introduces to simultaneously minimize costs and maximize service levels. Therefore, the overarching objective of this article is to establish optimal inventory levels for a variable periodic demand inventory model. In order to illustrate the the effectiveness of the proposed method, a numerical example is given.

Medical image feature extraction and selection based on InceptionV3 and Gini Index for cervical cancer cells identification

2026-01-18T04:33:13+08:00

Cervical cancer is one of the leading causes of death among women. The adverse effects of this cancer can be minimized with early diagnosis and treatment. In recent years, several machine learning models have been proposed for cervical cancer detection and prediction. In this paper, we evaluate a new framework that integrates feature embedding based on Inception v3 to detect cervical cell cancer from medical images, and use the Gini index to select the most informative features. The classification was executed employing k-Nearest Neighbors, Decision Tree, AdaBoost, Random Forest, and Artificial Neural Network algorithms. The implemented classifiers showed good accuracy results based on 9 and 5 selected features. The Random Forest algorithm outperformed the existent state-of-the-art research by achieving the best accuracy with only 5 features. These result show the efficiency of our model for the computer-assisted diagnosis and prevention of cervical cancer and could help physicians make diagnostic-based decision via pap-smear images.

Leveraging Ontologies and Process Mining in Personalized Recruitment Recommendations

2026-01-18T04:33:14+08:00

This paper presents a novel approach to improve recruitment methods through a comprehensive examination of contextual data and process models. The primary focus is on refining the process by aligning it with candidate preferences. The method incorporates ontology and process mining to provide contextual and sequential recommendations, adapting hunting methods according to candidate requests. Using a recruitment ontology and connecting it with candidate assessments, the approach refines strategies using successful recruitment historical data. Conformance checking identifies similar process models, connecting the ontology of each activity for a detailed analysis. The results highlight the effectiveness of the method in adjusting recruitment strategies based on historical and contextual data, offering a comprehensive and flexible solution for efficient recruitment.

Vision Transformers for Breast Cancer Mammographic Image Classification

2026-01-18T04:33:15+08:00

Background and Objective : The mortality rates due to breast cancer have been constantly growing and still represent one of the most common malignancies leading to death in females globally. Early and accurate detection is crucial to improve the survival rate. Recent deep learning advancements in artificial intelligence have opened a wide new avenue for further improving the results of computer-aided diagnosis. Vision transformers with their attention mechanism are among the recent promising ones, offering much-improved results for different image analysis applications, including mammography.

Methods : This study investigates the application of vision transformers and attention mechanisms for mammography image categorization. In this work, we used three publicly available datasets like the Mammographic Image Analysis Society (MIAS), Curated Breast Imaging Subset of DDSM (CBIS-DDSM), and INbreast. In the preprocessing of data, augmentation is used to enhance the generalization capabilities of models, and we have applied Contrast Limited Adaptive Histogram Equalization (CLAHE) to improve the contrast of images, especially in situations characterized by uneven lighting or low contrast levels.

Results : The proposed approach demonstrated superior performance compared to traditional convolutional neural network (CNN)-based methods. In the evaluation of this vision transformer, we have obtained an accuracy of 0.99, an AUC of 0.99 and an F1 score of 0.98.

Conclusion : Vision transformers and attention mechanisms have great potential to boost the detection of breast cancer using CAD systems. The findings accentuate their capability to improve the precision and reliability of mammography analysis, enabling early diagnosis and minimizing false positives and false negatives in clinical practice. The research emphasizes the need to embrace these new technologies to enhance patient outcomes and streamline healthcare resources.

A Mathematical Programming Model with Cost Sensitivity in the Objective Function for Imbalanced Datasets Challenges

2026-01-18T04:33:16+08:00

This paper introduces CS-MSD, a cost-sensitive deviation minimization model designed to address a typical issue of imbalanced datasets in binary classification, which is a major problem in machine learning tasks across various areas. Imbalanced datasets, in which a single class significantly dominates the other, frequently generate biased models that neglect the minority class, making them extremely important in practical sectors like health services and financial services. The Traditional re-sampling techniques, including under-sampling and over-sampling, have associated limitations, such as information loss and over-fitting. CS-MSD overcomes these limitations by combining external deviations with cost sensitivity, which produces a perfect balance of minority and majority class costs. The model outperforms Decision Tree and Radial SVM, achieving a Recall of 0.958 on the win dataset, alongside Specificity and G-mean metrics. With CPU and wall times of 0.052 s and 0.054 s, respectively, CS-MSD also surpasses Random Forest and Bagging in computational efficiency, making it ideal for time-sensitive tasks. Its combined flexibility and processing speed establish CS-MSD as a vital solution for enhancing classification performance across diverse domains.

Mathematical Modelling And Analysis Of Influenza (H5N1)

2026-01-18T04:33:17+08:00

In this study, we introduce a continuous MSEIHR model and explore its dynamic behavior and fundamental properties. Using Lyapunov functions and the Routh-Hurwitz conditions, we perform a stability analysis of the model. Our results confirm that when the basic reproduction ratio R0 < 1, the system is both globally and locally stable at the disease-free equilibria Eef . Reciprocally, when R0 > 1, an endemic equilibria Eeq emerges, and the system stabilizes at this equilibria. Additionally, we analyze the MSEIHR sensitivity to get the parameters with the most substantial influence on R0. Finally, we validate our theoretical findings with numerical simulations using Matlab.

Bingham type fluids with Tresca law in 3D: Existence, Asymptotic analysis, Reynolds equation

2026-01-18T04:33:18+08:00

In this work, we study a model for incompressible Bingham fluids in a confined three-dimensional domain, Ωε, where Tresca boundary conditions are applied on part of the boundary and Dirichlet conditions on another. The domain is perturbed by a small parameter ε > 0. We prove the unique solvability of the problem and carry out an asymptotic analysis as one dimension of the fluid domain diminishes to zero. This approach enables the strong convergence of the velocity field, the derivation of a Reynolds-type limit equation, and the analysis of the asymptotic behavior of the Tresca boundary conditions, while rigorously establishing the uniqueness of the limiting velocity and pressure fields.

Towards the application of Process Mining for analyzing Network Security

2026-01-18T04:33:19+08:00

The objective of our study is to investigate the use of process mining algorithms in analyzing data collected from a Man-in-the-Middle (MitM) attack that we simulated in a controlled local network to detect malicious activities. After analyzing the data we found results that indicate that under normal network conditions the BPMN model displays standard interactions without duplicate or conflicting IP addresses. However, in the case of MitM attack, the model shows duplicate and conflicting IP addresses which was a real manipulation and disruption. These results highlight how process mining can
improve forensic investigations into network security intrusions

A novel CNN architecture for breast cancer detection

2026-01-18T04:33:19+08:00

Breast cancer is a leading cause of mortality among women worldwide, and early detection is critical for improving survival rates. While mammography is a key screening tool, its accuracy can be impacted by human interpretation. Convolutional Neural Networks (CNNs) offer advanced image analysis capabilities to enhance early detection and support healthcare professionals with higher accuracy and reliability.

This study presents a novel CNN architecture, developed from scratch, to automate breast cancer detection and improve diagnostic accuracy. Using the MIAS and INBREAST datasets with advanced data augmentation techniques, the model demonstrates outstanding performance. On the MIAS dataset, it achieves an accuracy of 0.9912, recall of 0.9912, precision of 0.9914, AUC of 0.9996, and F1-score of 0.9912. Similarly, on the INBREAST dataset, the model achieves an accuracy of 0.9494, recall of 0.9494, precision of 0.9529, AUC of 0.9937, and F1-score of 0.9493, highlighting the accuracy and reliability across different datasets. The findings illustrate the potential of deep learning-based computer-aided diagnostic (CAD) systems in improving early breast cancer detection, reducing errors, and enhancing the cost-efficiency of providing healthcare.

Numerical modeling of natural convection in a square cavity filled with air: fractional derivative formalism

2026-01-18T04:33:20+08:00

This paper presents a numerical study of natural convection using the fractional derivative formalism. The model adopts nonlinear axis transformations and applies the finite difference method for spatial and temporal discretization in a square cavity filled with an incompressible fluid with a Prandtl number of $Pr = 0.71$. The configuration consists of four rigid walls, subject to a temperature gradient, which serves as the driving force behind the convection. No-slip and constant temperature conditions are applied on the walls. The governing equations are solved using fractional-order operators. Isotherms and streamlines are used to visualize the results, and the influence of varying the order of the fractional derivatives is analyzed to capture fine-scale flow and heat transfer features.

Optimized Dual Algorithm with Automatic Penalty Parameter Selection for Elastoplastic Contact Analysis

2026-01-18T04:33:21+08:00

In this article, We examine a mechanical contact problem involving an elastoplastic body and a rigid foundation. The behavior of the material is characterized by Hencky's nonlinear elastic constitutive law. We present an iterative method based on Kacanov's method, with an augmented Lagrangian formulation at each iteration. To improve the algorithm in the discrete case, we propose an alternative approach consisting of automatic and optimal selection of the penalty parameter, accompanied by an approximate algorithm. To this end, we eliminate two unknowns, the principal and the auxiliary, and thus formulate a purely dual algorithm, enabling us to study the convergence of our method in depth. Finally, numerical experiments on two-dimensional problems are conducted to demonstrate the performance of the proposed method.

Autoencoder-Based Reconstruction and Restoration of 3D Dental Objects

2026-01-18T04:33:22+08:00

This paper presents an exploration of autoencoders for 3D teeth reconstruction and completion, a crucial area in digital dentistry aimed at enhancing the efficiency of dental restoration and reconstruction. Accurate reconstruction of dental geometries is essential for developing personalized treatment plans and improving patient outcomes. However, most current approaches still rely on 2D imaging methods and often fall short of capturing the full complexity of tooth structures. In this study, we present a deep learning-based solution that effectively reconstructs 3D tooth models using point cloud representations. The results show that our method improve the accuracy of 3D tooth reconstruction and completion, as demonstrated by the results presented in these experiments. Our results have advancing digital dentistry techniques by providing a new methodology to utilize modern machine learning capabilities to enhance dental model reconstruction, which could lead to better treatment options, including the restorative treatments and the fabrication of customized dental prosthetics.

JetNet: An Effective Deep Learning Model for Histopathological Lung Cancer Classification and Diagnosis

2026-01-18T04:33:23+08:00

Effectively distinguishing lung cancer subtypes through histopathological imaging plays a vital role in accelerating diagnosis and guiding appropriate treatment strategies. Deep learning techniques, particularly convolutional neural networks (CNNs), have demonstrated remarkable success in medical image analysis. However, many state-of-the-art CNN architectures such as DenseNet, EfficientNet, and MobileNetV2 require substantial computational resources, limiting their clinical deployment in resource-constrained environments.

In this study, we propose \textbf{JetNet}, a novel CNN architecture designed to deliver both high classification accuracy and computational efficiency. JetNet incorporates a streamlined sequence of convolutional layers, batch normalization, global average pooling, and dropout regularization, resulting in a lightweight model with significantly fewer parameters. Evaluated on a publicly available histopathological lung cancer dataset, JetNet achieved an accuracy of 99.6%, outperforming well-established models including DenseNet, EfficientNet, and MobileNetV2.

The proposed model’s balance of performance and efficiency makes it particularly suitable for real-time diagnostic applications and deployment in clinical settings with limited computational infrastructure. This work advances automated lung cancer diagnosis and supports improved clinical decision-making.