Statistics, Optimization & Information Computing

Survival time in higher education program after a dropout using modified survival function: A retrospective study to predict average graduation time and factors leading to early dropout.

Intisar Siddiqui — 2025-11-22

Introduction:

The case recurrence survival model in terms of case retention after at least one dropout is still difficult to investigate, and some concrete framework is required to derive the survival model in order to achieve more precise results and reduce the magnitude of bias that may occur during dropout, reappearance, and retention either until the last or dropout again.

Design: Retrospective, longitudinal study

Place and duration:

The study was conducted in the School of Mathematical Sciences, College of Computing, Informatics, and Mathematics, Universiti Teknologi MARA, Malaysia, from December 15, 2023, to April 14, 2024.

Material and method:

Data for undergraduate program students spanning eight years was retrieved from the College of Dentistry, Imam Abdulrahman Bin Faisal University. It included students of any gender or any age group with at least 50% attendance in the first semester following the commencement of the program. Students who were expelled from the program based on violation or disciplinary action, deportation, imprisonment for criminal acts, or those with special needs or disabilities were excluded. Three possible events of survival function:

Retention: Attended the program in continuation till the end.
Dropout: Discontinuation of the program for a period of one or more semesters (or >6 months) with a zero-grade point average (GPA).
Retention after one dropout: Resumption of program after one dropout and retained in continuation till the end.

All relevant information was entered into the data worksheet of SPSS-29.0 (IBM product, USA). Syntax programming of the survival algorithm was developed using the statistical programming software R version 4.2.1, and survival parameters were generated.

Results:

The survival probability of the existing model compared to the modified function showed minimal differences. The survival rate was 94% in the first year of study, with a gradual decline of 1%–3% annually, reaching 91.6% by the end of the fifth year. The average survival time for the existing survival function was 4.666 ± 6.70 years, whereas the modified function exhibited a higher average of 5.584 ± 8.63 years. Similarly, the mean graduation time was slightly higher for the modified function (6.10 ± 0.302 years) compared to the existing model (6 ± 0.0 years).

Due to data confidentiality, only two variables were included as covariates in the Cox regression analysis: gender and reason for dropout. Among these, the reason for dropout was identified as a significant factor influencing student survival. Model performance, assessed using the R² value, indicated that the modified survival model was more accurate and preferable compared to the existing model (i.e., 0.903 vs. 0.808).

Conclusion:

It was concluded that dropout cases, which were censored in the existing survival model, played a significant role in estimating students’ survival time and the program’s graduation time. Hence, the modified function can be preferred when the first event to time doesn’t represent the final outcome.

Keywords:

Survival analysis, student dropout, higher education, retention, recurrence

A new family of bivariate alpha log power transformation model based on extreme shock models

Regent Retrospect Musekwa — 2025-11-24

This study presents a new bivariate distribution, referred to as the bivariate alpha log power transformation (BVALPT) model, developed by integrating the alpha log power transformation technique with the Marshall–Olkin extreme shock framework. Closed-form expressions for both the joint probability density function (pdf) and cumulative distribution function (cdf) are derived. The manuscript explores several key statistical properties of the proposed model, including marginal and conditional distributions, as well as survival and hazard rate functions. Parameter estimation is carried out using the maximum likelihood estimation (MLE) method. A notable special case, the bivariate alpha log power transformed exponential (BVALPTE) distribution, is examined in detail. The practical utility of the BVALPT family is demonstrated by fitting the BVALPTE distribution to a real-world dataset. Comparative results reveal that the BVALPTE offers an improved fit and enhanced analytical performance over the benchmark bivariate model considered in the analysis.

Proportionality test in the Cox model with correlated covariates

LAILI Lahcen — 2025-09-08

The correlation effect between covariates on the proportionality test results of a specific covariate in the Cox model is a problem that has already been reported by several authors. The first solution has been proposed for the Kolmogorov-Smirnov (KS) test, the Cramér-von Mises (CvM) test, and the Anderson-Darling (AD) test. It consists of simulating the null distribution of these test statistics, since this is only known if the covariates are uncorrelated. The results of the simulations carried out by the proponents of this solution have not proved its effectiveness in all studied cases. The second solution is based on the fact that the score function used in the tests mentioned above, and in the construction of the score tests, assumes that all other covariates are proportional, which is not always true. The idea is therefore to introduce temporal parameters to these covariates whose meanings match their proportionalities. Such a change in the score function requires estimation of the new parameters introduced for each tested covariate. In this article, we propose a simple technique to eliminate such an effect. The technique involves changing the covariate to be tested by the residual of its linear regression against the other covariates in the model. This change retains the same null hypothesis to be tested with a new covariate that is uncorrelated with the others. A simulation comparison of these techniques is considered.

Parameter and State Estimation in SIRD, SEIR, and SVEIR Epidemiological Models Using Kalman Filter and Genetic Algorithm

Didik Khusnul Arif — 2025-12-11

This study presents a comparative investigation of parameter and state estimation techniques applied to three epidemiological models: SIRD, SEIR, and SVEIR. The models are used to simulate infectious disease dynamics with increasing levels of complexity, incorporating factors such as exposure latency and vaccination. Parameter estimation is first performed using three approaches, they are Kalman Filter (KF), Extended Kalman Filter (EKF), and Genetic Algorithm (GA). The best parameter estimates from each method are then used as inputs for state estimation, which is carried out using the Extended Kalman Filter (EKF) and Unscented Kalman Filter (UKF). This forms six combinations of estimation strategies (KF-EKF, EKF-EKF, GA-EKF, KF-UKF, EKF-UKF, GA-UKF) evaluated across models. Root Mean Square Error (RMSE) is used as the evaluation metric to assess estimation accuracy. The results demonstrate that GA excels in estimating static parameters, while EKF is more effective for dynamic parameters. Hybrid combinations provide the best performance in state estimation across all models, indicating the benefit of combining global optimization and recursive filtering. These findings can support public health policy by informing the selection of appropriate modeling and estimation techniques to accurately predict epidemic trends, optimize vaccination strategies, and allocate medical resources more effectively during outbreaks. All experiments are conducted on synthetically generated epidemic data to ensure controlled evaluation and generalizability across models.

Hybrid Robust Beta Regression Based on Support Vector Machines and Iterative Reweighted Least Squares

Taha Hussein Ali — 2025-12-12

In this paper, we examine and compare the performance of several beta regression approaches for response variables constrained to the (0,1) interval, focusing on robustness in the presence of outliers and nonlinear relationships. Since the beta distribution is well suited for modeling proportions, it is used here to describe the rate of tumor response to cancer therapy. Four modeling strategies are considered: standard beta regression estimated via maximum likelihood; robust beta regression using the IRLS-Huber procedure; support vector regression (SVR) followed by a beta transformation; and a hybrid beta regression model that combines SVR with Huber-based robustness. The models are assessed using a simulated dataset generated under controlled levels of contamination and varying sample sizes, as well as a quasi-real tumor response dataset in which age is the primary covariate. The simulation results indicate that although classical least squares (CLS) and robust beta regression can provide adequate predictions under ideal conditions, their performance deteriorates when outliers are present and the relationship is nonlinear. While SVR better captures nonlinear patterns and therefore outperforms the other individual methods, it also lacks robustness to contaminated data. Across all conditions, the hybrid model achieves higher accuracy and greater robustness, reflecting strong generalization capability and adaptability. When applied to the real tumor response data, the hybrid method again emerges as the preferred model, effectively accommodating outliers and delivering the most stable and precise predictions. Overall, the hybrid SVR-Huber beta regression framework proves to be a valuable and powerful tool for medical research and other applied fields that must analyze noisy, bounded real-world data.

Efficient Randomized Response Model Tailored for Estimating Highly Sensitive Characteristics

Ahmad Aboalkhair — 2025-09-26

When broaching extremely delicate subjects, individuals might offer inadequate or dishonest revelations, jeopardizing data precision. To counteract this challenge, this research proposes a new and effective randomized response structure crafted to enhance the assessment of highly sensitive characteristics. The proposed framework enhances Aboalkhair’s (2025) model, which has emerged as a viable substitute for Mangat’s frameworks. This investigation assesses the scenarios where the proposed method performs better than Mangat's method. Through theoretical scrutiny and numerical simulations—taking into consideration partial honest disclosures—the outcomes showcase the model's heightened effectiveness. Furthermore, the article quantifies the level of privacy safeguarding provided by this innovative approach.

The Extreme Value Theory for Demographical Risk Analysis and Assessment: Peaks Over Random Threshold Value-at-Risk Analysis of Regional Prevalence Data with a Demographical Case Study

Nazar Ali Ahmed — 2025-10-17

This study presents a novel application of Extreme Value Theory (EVT) to analyze and quantify the risk associated with disability prev lence across regions in Saudi Arabia. Leveraging advanced actuarial risk modeling techniques including Value-at-Risk (VaR), Tail Value-at-Risk (TVaR) , Mean of Order P (MOP), and the Peaks Over Random Threshold Value-at-Risk (PORT-VaR) framework, we provide a robust statistical assessment of regional disparities and extreme disability risk. Using demographic data from the 2016 national survey, our analysis identifies critical outliers and quantifies extreme thresholds at varying confidence levels (55%–95%). The Northern Border region emerges as a high-risk area, with over 285,000 individuals living with disabilities, significantly exceeding other regions. Our PORT-VaR and Peaks Over Random Threshold Mean of Order P (PORT-MOP) models highlight urgent policy targets and resource allocation needs, particularly for older adults and low-income populations who are disproportionately affected. This work contributes to the growing field of actuarial statistical modeling by demonstrating how EVT-based tools can enhance public health planning and support evidence-based decision-making in social policy development. By aligning traditional actuarial methodologies with contemporary public health challenges, the study underscores the relevance of predictive modeling and quantitative risk management in addressing complex societal issues such as disability prevalence.

Bayesian estimation, Bayesian neural network and maximum likelyhood estimation for a novel transmuted tangent family of distributions with applications in healthcare data

Amrutha P T — 2025-11-17

Cancer is currently the main cause of primary or secondary premature mortality in most countries. Medical researchers require statistical analysis to identify the most suitable model for assessing the remission periods or survival times of cancer patients, thereby producing precise results. The current study contributes a novel family of distributions to analyze the remission periods or survival times of cancer data effectively, termed the transmuted tangent family of distributions, achieved through the combination of the quadratic transmuted family with the tan-G class of distributions. The primary statistical properties of the proposed family are established. The Bayesian estimation, Bayesian neural network and maximum likelihood estimation methods are employed for parametric estimation of the family. In addition, four members of the family are introduced. The transmuted tangent Lindley distribution is examined, and its fundamental features are established. Three cancer datasets are examined to verify the fit efficiency of the proposed family through the use of various goodness-of-fit measures. We have demonstrated that, compared to many established families, the proposed family offers a better fit to the data sets.

Cyclical properties of Moroccan migrant remittances: an empirical analysis

SAMIR FARHI — 2025-12-03

Remittances constitute a major source of financing for most developing countries, due to their growing volume and their potential contribution to the growth and development of the countries of origin. Morocco is no exception to this dynamic. The literature on the cyclicality of these flows reveals that they can be procyclical or countercyclical. This cyclicality has major implications for economic and financial policies. A better understanding of these dynamics would make it possible to optimize the use of these funds in times of expansion as well as in times of crisis, thus strengthening their role as a stabilizer and shock absorber. In this context, this study analyzes the cyclicality of remittances to Morocco during the period 1980-2022, using appropriate econometric filters and a vector autoregressive model (VAR) incorporating impulse response functions (IRF), in order to study the interactions between these flows and economic cycles. This article argues that the cyclical nature of remittances must be assessed in a dynamic framework.

Reliability Modeling with a Transformed Exponential Distribution: Theory and Applications

Naser Odat — 2025-11-18

In this study, we present the Transmuted Exponential Distribution (TED), which is a quadratic transformation-based version of the exponential distribution. The TED gets around the exponential distribution's main drawback, which is its constant failure rate, by introducing a single shape parameter . The new model is much more appropriate to data from the real world since it can depict both rising and falling hazard rates. Using a real-world dataset, we show its greater performance over the conventional exponential model and deduce its fundamental statistical features.

Efficient Test for Threshold Regression Models in Short Panel Data

DOUNIA BOURZIK — 2025-12-15

In this paper, we propose locally and asymptotically optimal tests (as defined in the Le Cam sense) that are parametric, Gaussian, and adaptive. These tests aim to address the problem of testing the classical regression model against the threshold regression model in short panel data, where n is large and T is small. The foundation of these tests is the Local Asymptotic Normality (LAN) property. We derive the asymptotic relative efficiencies of these tests, specifically in comparison to the Gaussian parametric tests. The results demonstrate that the adaptive tests exhibit higher asymptotic power than the Gaussian tests. Additionally, we conduct simulation studies and analyze real data to evaluate the performance of the suggested tests, and the results confirm their excellent performance.

Choosing Best Robust Estimation Parameters among Some Exponential Family Distributions

Hozan Taha Abdalla — 2025-11-26

Survival analysis is statistical route deliberated to explore the time until an event appears. Most research on the reliability of the survival function has some shortcomings in the process of accurate statistical analysis that aims to obtain highly efficient estimates. The necessity for these efficient estimation methods is called robust methods which becomes important when the data for the phenomenon being studied is contaminated, i.e., there are outliers in the observations, which results in estimates that lead to an increase or decrease in the mean square error (MSE), which leads to inaccurate statistical inference. A breast cancer is commences from the abnormal growth of the cells in the mammary gland, ductal carcinoma or lobular carcinoma of the breast the data that has been used has 4024 observations and the independent variables are (Age, Tumor Size, Tumor Stage, and Node Stage). The study focusses on comparing different methods such as (AFT), Regression model, Robust AFT with Tukey’s Biweight function, Robust AFT with Median function and Robust Regression model with Tukey’s Biweight function for each of (Exponential, Weibull and Lognormal) distributions and non-parametric bootstrap is used to derive the standard error, z-values and p-values for each of the classical and robust methods, ensuring robust inference free from asymptotic assumptions.
The Aim of this study is to choose the best method from each of (AFT), Regression model, Robust AFT with Tukey’s Biweight function, Robust AFT with Median function and Robust Regression model with Tukey’s Biweight function for each of (Exponential, Weibull and Lognormal) distributions and to choose those parameters that affect the survival time. A comparison was conducted among AFT, Regression model, Robust AFT with Tukey’s Biweight function, Robust AFT with Median function and Robust Regression model with Tukey’s Biweight function. We conclude that that the Robust AFT Lognormal Distribution in Survival Analysis by using Maximum likelihood-Type Estimator with Tukey's Biweight function is the best method, the Robust AFT Weibull Distribution in Survival Analysis by using Maximum likelihood-Type Estimator with Tukey's Biweight function is also the best model. As well as the Robust AFT for (Exponential, Weibull and Lognormal) by using Maximum likelihood-Type Estimator with Median weight function shows the best result.

Improving Set-union knapsack problem based on binary spotted hyena optimization algorithm

Rana Hussein — 2025-10-24

One pertinent model for intelligent systems and decision making is the Set-union Knapsack Problem (SUKP). Heuristic algorithms are helpful in finding high-quality answers in a reasonable amount of time, despite their inherent difficulty (NP-hardness). The binary spotted hyena optimization algorithm for the set-union knapsack problem is presented in this study. Numerous heuristic and approximation techniques for resolving the set-union knapsack issue have been documented in the literature. The quality of the solution still has to be improved, though. The purpose of this study is to apply Z-shaped transfer functions to the binary spotted hyena optimization algorithm used to solve the Set-union knapsack problem. Comparative experimental results show that Z-shaped transfer functions are competitive or superior than the other state-of-the-art transfer function. The experiments were done on three types of 30 popular SUKP benchmark examples.

Selection of Initial Points Using Latin Hypercube Sampling for Active Learning

Nompilo Mabaso — 2025-11-24

Classification requires labelling large sets of data, which is often a time-consuming and expensive process. Active learning is a machine learning technique that has gained popularity in recent years due to its ability to effectively reduce the amount of labelled data required to train accurate models. The success of the active learner heavily relies on the selection of the initial points to initialise the active learning process. In this paper, we compare the performance of the traditional random sampling approach to the maximin Latin Hypercube sampling, conditioned Latin Hypercube sampling, and a modified Latin Hypercube sampling procedure for initialising active learning for the estimation of the logistic regression in binary classification problems. We show that the Latin Hypercube sampling designs outperform random sampling for all the performance measures evaluated. The results are demonstrated using simulated data sets and an actual case study. Specifically, the conditioned Latin hypercube sampling design exhibits high prediction accuracy using a smaller sample size for both heterogeneous and homogeneous classes. In contrast, the modified Latin hypercube sampling design yields the smallest variance of prediction across varying initial sample sizes for both homogeneous and heterogeneous classes. Furthermore, principal component analysis indicates that approximately 10\% of the data is required to develop an accurate and precise logistic regression classifier.

The Central Metric Dimension of the 𝒌-Corona Graph

Liliek Susilowati — 2025-11-17

The metric dimension is the minimum cardinality of a subset of the vertex set of a graph that uniquely represents each vertex in a graph. The central set is a set of vertices with minimum eccentricity. This central set concept can be used to determine strategic public service locations, such that accessible transportation can be reached from all regions. The central metric dimension is the minimum cardinality of a resolving set that includes the central set. This study aims to determine the central metric dimension in k-corona graph. The k-corona operation of G and H denoted by GoH is a generalization of the corona operation, where a new graph is formed by connecting each vertex of a graph G to k copies of graph H. The results show that the central metric dimension of the k-corona graph depends on the central set of G , the order of G , the value of k, and the metric dimension of H .

The sensitivity analysis of multistate pension projections based on a vec-permutation approach

Nada El Moutaki — 2025-12-08

The cohort component method in the context of pension projections can be translated into a multistate matrix model, in which beneficiaries of a pension scheme are classified jointly by their age and status (active contributor, invalid, retiree, widows/widowers), using the vec permutation matrix. The projection results depend on the mortality, retirement, disability, marital percentage and remarriage rates as well as the number of new entrants into the scheme on which the projections are based. Any change to a parameter will result in a corresponding change in the projection outcomes. Our objective is to systematically examine the relationships between various key projection outcomes—such as status-specific population sizes, dependency ratio, total
cash flows, and PAYG cost rate—and the underlying age- and sex-specific projection parameters. To achieve this, we present the set of equations required to perform sensitivity and elasticity (i.e., proportional sensitivity) analyses of multistate projections, utilizing matrix calculus. We apply our methodology to a projection of the Moroccan pension system, which estimates population and cash flows disaggregated by age, sex, and status over the period 2020 to 2080.

Contingency Uniformity Measure: An approach for spread characterization in Contingency tables

Ratnesh Kumar Singh — 2025-12-12

This paper introduces the Contingency Uniformity Measure (CUM), a normalized entropy that scales Shannon entropy to the range [0, 1], enabling fair comparisons across contingency tables of different dimensions. CUM retains key properties such as nonegativity, reaches its maximum at the uniform distribution, and satisfies weighted additivity. We formulate and solve three optimization problems, using CUM, under realistic constraints, fixed marginal distributions, a cost matrix, and cost variance. This is demonstrated through a real dataset of cost matrix obtained using distance matrix. Our results show that CUM is an effective, standardized measure for analyzing uncertainty and supporting decision-making in diverse, constraint-driven systems.

Novel Hybrid Conjugate Gradient Technique Based on the Newton Direction Applied to Image Restoration Problem

Romaissa Mellal — 2025-11-12

We introduce a novel hybrid conjugate gradient method for unconstrained optimization, combining the AlBayati-AlAssady and Rivaie-Mustafa-Ismail-Leong approaches, where the convex combination parameter is determined to ensure alignment between the conjugate gradient direction and the Newton direction.
Through rigorous theoretical analysis, we establish that the proposed method guarantees sufficient descent properties and achieves global convergence under the strong Wolfe line search conditions.
Numerical experiments on image restoration confirm that our method exhibits competitive or superior performance compared to the Fletcher-Reeves algorithm, especially when processing images with higher noise levels.

A Stochastic Optimization Model for a Multi-Echelon Inventory System with Direct Demand that Consists of Two Commodities

PRABAHARAN K — 2025-12-08

In this paper, we present a stochastic optimization model for a multi-echelon inventory system with direct
demand, handling two interrelated commodities. The system consists of a three-level continuous review inventory model,
comprising a warehouse (WH), a single distribution center (DC), and a retailer (R). A (s; S) inventory policy is implemented,
assuming Poisson demand and exponentially distributed lead times at the retail node. The DC replenishes retailers in fixed
pack sizes Qi(= Si - si), while the WH provides an abundant supply. We derive the steady-state probability distribution
and key performance measures, offering insights into system efficiency and operational characteristics

Local Total Distance Irregularity Labeling of Graph

Eric Dwi Putra — 2025-11-09

We introduce the notion of total distance irregular labeling, called the local total distance irregular labeling. All edges and vertices are labeled with positive integers 1 to k such that the weight calculated at the vertices induces a vertex coloring if two adjacent vertices has different weight. The weight of a vertex $u\in V(G)$ is defined as the sum of the labels of all vertices adjacent and edges incident to $u$ (distance $1$ from $u$). The minimum cardinality of the largest label over all such irregular assignment is called the local total distance irregularity strength, denoted by $tdis_l(G)$. In this paper, we established the lower bound of the local total distance irregularity strength of graphs $G$ and determined exact values of some classes of graphs namely path, cycle, star, bipartite complete, fan and sun graph.

Trigonometrics Functions Algorithm : a novel metaheuristic algorithm for engineering problems

Wendinda Bamogo — 2025-12-08

This paper deals with the design of a novel metaheuristic algorithm called Trigonometrics Functions Algorithm
(TFA) for efficient solving of engineering problems. The fundamental inspiration for this new algorithm is based on a
mathematical model inspired by the hunting and attack technique of grey wolves and using trigonometrics functions. For
better exploration and exploitation of the search space, several random and adaptive variables are used. The various stages
of well-arranged TFA are described and mathematically modeled. In order to prove the effectiveness and robustness of
TFA, many engineering optimization problems of different difficulties were solved and a statistical study was made. The
optimization results obtained with TFA were compared with the results of other state-of-the-art algorithms. Statistical and
comparative studies showed that TFA achieves the best results and generally ranks first among the solved problems. The
study of the sensitivity of TFA related to several parameters shows that TFA has a high degree of stability giving it the ability
to efficiently solve optimization problems. In summary, the various studies have highlighted the efficiency, robustness and
superiority of TFA compared to other competing algorithms and thus allow us to conclude that TFA remains a better option
for solving technical design optimization problems.

Results on Grundy Chromatic Number of Prism graphs

Annathurai K — 2025-12-17

A coloring of a graph G is a proper vertex coloring of G having the property that for every two colors i and j with i < j, every vertex colored j has a neighbor colored i. we acquire the Grundy chromatic number of prism, Crossed Prism graph, Antiprism graph and the Line graph of Crossed Prism graph with suitable illustrations whenever necessary.

The Log Akash Regression Model with Application

Samy Mohamed — 2025-12-15

The current study proposes and presents a new regression model for the response variable following the Akash
distribution. The unknown parameters of the regression model are estimated using the maximum likelihood method. A simulation study is conducted to evaluate the performance of the maximum likelihood estimates (MLEs). Additionally, a residual analysis is performed for the proposed regression model. The log-Akash model is compared to several other models, including Weibull regression and gamma regression, using various statistical criteria. The results show that the suggested model fits the data better than these other models. It is anticipated that the model has applications in fields such as economics,biological studies, mortality and recovery rates, health, hazards, measuring sciences, medicine, and engineering.

Impact of Fear, Allee Effects, and Harvesting on a Predator-Prey Delay Model with a Modified Beddington–DeAngelis Functional Response

Miswanto — 2025-11-02

This paper studies the dynamics of a delayed predator-prey model with a modified Beddington-DeAngelis response function influenced by fear factors, Allee effects, and harvesting on the predator population. This paper analyzes the influence of parameters, namely fear factors ($\omega$), Allee effects ($m$), and delay time ($\tau$), on the stability of the model’s equilibrium point. First, an analysis of the existence of the model’s equilibrium point is carried out, then an analysis of the stability and the influence of changes in the model’s parameter values and delay time that can affect the stability of the model’s equilibrium point is carried out. The analysis indicates that the larger the parameters $\omega$, $m$, and $\tau$, the more unstable the coexistence equilibrium point tends to be. Several numerical simulation results are used to validate the analytical results obtained.

A Theoretical Mathematical Model for Exploring Learning Dynamics: A Scenario-Based Analysis

J.Igbida — 2025-12-10

This work offers a theoretical reflection on interactions between students and teachers through the development of four modelled scenarios. Using mathematical tools, it seeks to analyse, in an abstract manner, the dynamics of learning and how they evolve in response to different forms of pedagogical intervention. Each scenario is designed as a conceptual illustration highlighting essential dimensions of teaching and learning, such as feedback, adaptability and personalisation. The aim is not so much to study actual practices as to show, through modelling, how these mechanisms can be understood and interpreted. This work thus aims to enrich the theoretical understanding of educational processes and highlights the contribution of formal models as a framework for analysing the complexity of pedagogical interactions.

Lung Cancer Segmentation and Classification with Multi-Dataset Integration

Hozan Abdulqader — 2025-09-28

Accurate computer-aided lung cancer diagnosis is based on two sequential tasks: precise nodule segmentation and reliable malignancy classification. To this end, we curated the largest open-source CT benchmark to date by unifying five public repositories, resulting in 7,061 annotated slices from 571 patients for segmentation and 17,351 slices from 1,208 patients for classification. A standardized pre-processing pipeline was developed to harmonize voxel spacing, intensity windows, and label conventions.
For segmentation, six encoder–decoder architectures were evaluated, with the hybrid UNet++ achieving the highest validation performance (Dice coefficient = 98.5%), demonstrating that attention-augmented dense skip pathways enable more accurate boundary detection of lung nodules.
These masks were then used to drive a two-phase classification strategy: models were initially trained using ground-truth masks, followed by fine-tuning on predicted masks to emulate real-world deployment scenarios. Our proposed NoduleHyperFusionNet a dual-stream EfficientNetV2-S architecture , achieved the best overall discrimination (Accuracy = 92%, F1-score = 89%, AUC = 91%). The EfficientNet-B3 model also performed strongly, reaching an AUC of 94%.

Overall, this study demonstrates that the combination of attention-enhanced segmentation and lightweight multichannel fusion architectures can significantly improve automated lung cancer workflows, reducing diagnostic error rates without incurring prohibitive computational costs.

Efficient Power Flow Solution in Monopolar DC Networks Using a Derivative-Free Steffensen Method

Oscar Danilo Montoya Giraldo — 2025-10-01

This paper proposes and evaluates advanced solution techniques for nonlinear power flow analysis in monopolar low-voltage DC networks. The key contribution is the systematic comparison between the classical Newton-Raphson method and a derivative-free multivariable Steffensen method, demonstrating that the latter offers a practical alternative with superlinear convergence, reduced computational complexity, and simpler implementation. Numerical simulations on benchmark 33-bus and 69-bus systems show that both methods converge rapidly within fewer than six iterations, with Steffensen’s method maintaining competitive solution times and accuracy while significantly lowering the effort needed for Jacobian evaluations. The findings confirm that the Steffensen method is highly suitable for real-time, large-scale power system analysis, especially where derivative calculations are expensive or unreliable. Overall, the results endorse the Steffensen approach as a robust, efficient, and scalable solution method for modern DC power systems, paving the way for improved operational reliability and integration of renewable energy sources.

Securing IoT Systems Using Artificial Intelligence-Driven Approaches

Obaida M. Al-Hazaimeh — 2025-12-25

The Internet of Things (IoT) has transformed modern infrastructure by connecting billions of smart devices, yet faces critical security challenges due to restricted processing power, diverse communication protocols, and delayed security implementations. Traditional cybersecurity approaches and conventional deep learning methods inadequately address these threats while maintaining computational efficiency for resource-constrained IoT environments. This paper presents a novel hybrid framework combining Discrete Orthogonal Hahn Moments with EfficientNet deep learning architecture for enhanced IoT attack detection. The methodology leverages Hahn Moments' superior feature extraction through weighted orthogonality properties to reduce dimensionality while preserving discriminative information. Integration with EfficientNet-B0's compound scaling and Mobile Inverted Bottleneck Convolution blocks enables efficient learning with only 5.3 million parameters a 77% reduction compared to traditional networks. Experimental validation demonstrates remarkable performance, achieving 99.6% detection accuracy with 99.63% specificity and 98.99% sensitivity at 232×232 resolution. The proposed framework outperforms K-nearest network (84.6%), Multiple Linear Regression (88.2%), Parse Tree (93.7%), Latent Semantic Analysis (97.9%), and conventional Deep Neural Networks (98%) while maintaining minimal computational overhead of 38 seconds. Results establish this hybrid approach as a robust solution for real-time IoT security monitoring in resource-constrained environments.

A Novel Deep Learning Technique for Big Data Anomaly Threat Severity Prediction in ELearning

Chinnakka Sudha — 2025-12-08

E-learning platforms are susceptible to several anomalies, including abnormal learning behaviours, system abuse, and cyber security (CS) attacks, which interfere with the learning process. Conventional methods for detecting anomalies have limitations with high-dimensional data, skewed distributions, and poor feature selection, resulting in incorrect severity level predictions. To overcome these, a novel Sea Lion Multilayer Perceptron (SLMP) model is introduced for anomaly severity level prediction. At First, an e-learning anomaly dataset is gathered and trained in a Python environment. Hence, the data is preprocessed, and the Sea lion optimization (SLO) is used to select the best features to attain only the most significant attributes. Subsequently, the chosen informative features are employed for further process. Moreover, prediction and classification are performed using the SLMP model. Finally, Performance metrics like F score, Accuracy, recall precision and error rate are used to evaluate the effectiveness of the model. The results confirm the efficacy of the developed SLMP framework over current methods, illustrating its strength in optimizing predictive efficiency for anomaly severity detection in e-learning systems.

Economic order quantity model with green technology approach considering time dependent demand in fuzzy environment

R. Janani — 2025-11-02

In this study, we consider the problem of determining the effect of carbon tax and green technologies over infinite planning horizon. We have developed two economic order quantity models considering constant demand and exponential time dependent demand. Carbon emissions may also occur from storing undelivered or unsold products due to some factors. Investment in green technology is also considered. In this article, we fuzzify the inventory parameters such as demand, ordering cost, holding cost and the amount of carbon emitted when storing the products. To incorporate uncertainty, pentagonal fuzzy numbers are utilized to fuzzify the parameters of the inventory system. The graded mean integration method is used for defuzzification. The optimal values of the order quantity, cycle time and optimal inventory cost in both crisp and fuzzy sense is determined for both the models. Numerical illustrations are given to demonstrate the solution procedure and the sensitivity of various parameters are analysed.

On Fixed Points of Tirado-Type Contractions in k-Neutrosophic Metric Spaces

N. Muthulakshmi — 2025-10-15

The notion of a generalized k-Neutrosophic Metric Spaces and its inherent characteristics are presented in this study. A new class of contraction mappings, termed Tirado-type k-neutrosophic contraction mappings, is defined, extending classical contraction principles to this novel framework. A fixed-point theorem is established under the G-completeness condition, and an illustrative example demonstrates the generalized naturalness of k-neutrosophic metric spaces. These findings bridge this gap, unifying and extending fixed-point theories in more generalized settings.

EVEN AND ODD ANTIMAGIC TOTAL LABELING ON TOTAL GRAPH OF PATHS AND CYCLES

N. Nagaparvatham — 2025-11-02

Consider a graph G = (V,E) with s nodes and t links. In this paper, we introduce two types of labeling, namely odd-vertex(edge) antimagic total labeling and even-vertex(edge) antimagic total labeling. A vertex(edge) antimagic total labeling is called odd vertex(edge) antimagic total labeling if φ(V (G)) = {1, 3, ..., 2s − 1} and it is even vertex(edge) antimagic total labeling if φ(V (G)) = {2, 4, · · · , 2s}. Here we discuss even and odd vertex antimagic total labeling and edge antimagic total labeling on total graph of paths and cycles.

Bipolar Valued Vague Subfields of A Field By Mapping

K. Bala Bavithra — 2025-11-15

In this paper, we introduce functions defined on the bipolar-valued vague subfield of a field and investigate their fundamental properties. Since functions play a crucial role in the study of bipolar-valued vague subfields, particular attention is given to the application of homomorphisms within this framework. Furthermore, one or two types of translations are employed in this work, serving as tools to establish communication and relationships between different structures.

Optimized Deep Ensemble Framework for Colorectal Polyp Detection and Clinical Deployment Design

Fayza Elshorbagy — 2025-12-21

Early and accurate detection of colorectal polyps is critical for reducing colorectal cancer risk and improving patient outcomes. This paper introduces an ensemble deep transfer learning framework with Bayesian hyperparameter optimization for robust colorectal polyp classification. The method combines three state-of-the-art backbones—ResNet50, EfficientNetB0, and InceptionV3—whose outputs are fused via probability averaging to improve reliability. Stratified 10-fold cross-validation provides unbiased performance estimates, while Bayesian optimization fine-tunes model parameters for high accuracy and efficiency. Experiments on two benchmark datasets demonstrate excellent results, achieving 99.56% accuracy on CP-CHILD-A and 99.40% on CP-CHILD-B. To illustrate clinical usability, we also designed user interface prototypes as a computer-aided diagnostic (CAD) system, showing how the framework could be integrated into real-world screening workflows. These results highlight the potential of the proposed approach for real-time, clinically deployable colorectal polyp detection.

Cosets and Congruence Relations over Residuated Lattices in the SoftMultiset Framework

ANUSUYA ILAMATHI VS — 2025-11-20

This paper introduces cosets, congruence relations, and equivalence relations of softmultiset filters over residuated lattices. Various properties of
these structures are examined, providing insights into their algebraic char-
acteristics and foundational role in soft-multiset theory.

CancerSeg-XA: Medical Histopathology Segmentation System Based on Xception Backbone and Attention Mechanisms

Alaa Youssef — 2025-12-03

Accurate segmentation of histopathological images is essential to support early diagnosis and effective treatment planning in cancer care. This study presents CancerSeg-XA, a deep learning-based histopathology segmentation system designed to deliver robust performance across diverse tissue types and imaging sources. Built upon the DeepLabV3+ framework, CancerSeg-XA incorporates architectural enhancements to strengthen feature representation and improve model stability. The system was evaluated on three widely recognized datasets—BCSS, PanNuke, and PUMA—each presenting distinct structural and clinical challenges. Across all datasets, CancerSeg-XA consistently outperformed the baseline DeepLabV3+ in terms of segmentation accuracy, recall, and F1-score. Specifically, it achieved accuracy improvements of 4.78%, 4.31%, and 3.22% on BCSS, PanNuke, and PUMA, respectively, along with substantial gains in FwIoU. These results highlight the model’s ability to generalize effectively across varied histopathological contexts, positioning CancerSeg-XA as a promising solution for clinical integration and future research in digital pathology.

Cross-Modal Federated Learning for Robust Plant Disease Classification

Souad LAHRACHE — 2025-11-24

The accuracy of automated plant disease diagnosis is frequently limited by the use of visual symptoms alone, especially when it comes to differentiating between conditions that have a lot of visual similarities. To address this, we propose a new privacy-preserving framework that combines the strengths of multi-modal federated learning (FL) with environmental context. Our system integrates leaf images with synthetic sensor data—such as temperature, humidity, and leaf wetness duration—capturing critical cues that influence disease progression. Actually, this system core is a dualbranch convolutional neural network designed to process both image and environmental features in a way that reflects the biological characteristics of different diseases. Results demonstrates that the multi-modal approach consistently outperforms conventional image-only models across multiple disease categories, and especially true for diseases where environmental factors are very important in how they develop. We further extend the system into a federated learning setting, allowing models to benefit from distributed training while keeping sensitive agricultural data local and private. This makes the framework not only more accurate but also practical for real-world use, where data privacy is essential.

Diagnosis with Deep Learning and a Novel Pre-processing Strategy on a Large-Scale Dermoscopic Image Dataset

Youssra EL IDRISSI EL-BOUZAIDI — 2025-12-08

Skin cancer, in particular melanoma, has become a major health problem worldwide. Early diagnosis is the most important factor to consider for successful treatment. The latest advances in diagnosis have increased melanoma survival rates significantly, include early detection techniques such as imaging technologies that can detect melanoma in its earliest stages when treatment is most effective. AI has also been a game changer in the field of diagnosis, providing automated analysis data with a high level of accuracy. In this article, a novel computer-assisted diagnosis is presented, which consists of a new preprocessing technique to improve the quality of images. Data augmentation is used to increase data size by applying transformations to improve model generalization. The transfer learning efficiency is proved using the MobileNetV2 model. Improving and fine-tuning this architecture for the skin lesion classification task. The trained model can achieve performance, with an accuracy of 95.1% on 7 classes, and a very high AUC score of 94% for the precision-recall curve on the HAM10000
benchmark dataset. These results show how advanced deep learning techniques can be used in dermatological practice, thus creating a promising alley for improved skin cancer diagnosis.

A Lambda Lakehouse Architecture Bridging Streaming and Batch Intelligence in Volatile and Scalable Financial Data Processing

Maryam Maatallah — 2025-12-21

The vast growth of digital financial market data necessitates new kinds of analytical infrastructure which can process large volumes of data continuously, while maintaining reliability for use over extended periods as part of a long-term historical processing requirement. Batch based platforms have difficulty meeting both these needs, whereas pure streaming platforms often sacrifice analytical consistency with respect to their analysis.

To address this limitation our paper proposes a Unified Lambda-Lakehouse Architecture which allows Real-Time and Batch Processing to be performed together in a single, ACID compliant. Apache Kafka captures live Bitcoin markets and performs the real-time processing via Spark Structured Streaming, while the periodic storage of historical records and subsequent periodic reprocessing of those records is accomplished via Amazon S3. Ultimately both the real-time and batch processing paths converge at a Delta Lakehouse; thereby enabling schema enforcement, versioning, and time-travel queries.

The proposed architecture places the emphasis on combining the Speed Layer, Batch Layer, and Serving Layer into a single operational workflow atop a transactional Lakehouse foundation. Advanced predictive models including LSTM, GRU, ARNN, and XGBoost are used to forecast Bitcoin prices at daily, hourly, and minute granularities. Results from experiments indicate that the LSTM model consistently produced the best results (RMSE = 2383.9, 539.3, 144.9) at the three respective levels.

Physics-Informed Transformer Networks for Multi-Peril Insurance Pricing: A Novel Hybrid Computational Framework Integrating Actuarial Principles with Deep Attention Mechanisms

Eslam Seyam — 2025-12-12

A classic problem in insurance pricing is the trade-off between actuarial validity and predictive accuracy. Traditional Generalized Linear Models follow insurance principles rigorously but don’t forecast very well, while machine learning algorithms offer strong predictive performance but ignore key insurance rules. To close this gap, we extend the Transformer architecture into what we call a Physics-Informed Transformer. This model integrates five core insurance rules—premium adequacy, monotonicity, multiplicative decomposition, calibration, and coherence—directly into both the architecture and the loss function. The proposed Physics-Informed Transformer uses multi-head attention to learn non-linear relationships among features while maintaining actuarial validity through a combination of soft and hard constraints. We tested the model on French motor insurance data with 108,699 samples. The results show that the Physics-Informed Transformer outperforms existing models in actuarial performance, achieving a Gamma deviance of 1.0756. It also reaches over 99% compliance with insurance validity rules, whereas typical machine learning algorithms do not comply with these rules. To quantify how well the model adheres to insurance rules, we introduce a new measure called the Actuarial Validity Score (AVS). The proposed model achieves an AVS of 0.7659, representing a 23% improvement over traditional GLM models and matching the performance of Gradient Boosting Machines. Beyond prediction accuracy and rule compliance, the model’s attention mechanism highlights actuarially meaningful feature interactions without the need for manual feature engineering, offering valuable insights that support regulatory approval and acceptance. However, despite its advantages, the Physics-Informed Transformer still falls short in certain areas, particularly in Calibration Retention (10% compliance) and Monotonicity Preservation (72% compliance).

Design of an Iterative Multi-Analytical Fuzzy TOPSIS Framework for Enhancing Electoral Decision Systems in India: Interpretability, Regional Adaptation, and Policy Simulations

Kiran Kumar Chanumolu — 2025-12-22

The integrity, transparency, and inclusivity of the voting systems in India need to be empowered by the diverse demographics of the nation and the complex manner in which socio-economic and political factors interact towards influencing electoral behavior. Traditional voting analysis frameworks often rely on rigid statistical models that fail to capture the ambiguity inherent in human decision-making, especially in terms of subjective judgments and linguistic assessments. Existing models, therefore, remain unfit for purposes of policy-level decision support, owing to factors of lack of temporal adaptability, regional granularity, and scalable validation mechanisms. To remedy these limitations, this study proposes the development of a novel multi-methodological framework based on Fuzzy TOPSIS, augmented with five novel analytical extensions for model implementation and validation. The Explainable Causal Inference Layer integrated with Fuzzy TOPSIS (XCI-FTOPSIS) stands for traceable and interpretable prioritization of preferences by voters. The Spatiotemporal Attention-Based Fuzzy Decision Matrix (SA-FDM) captures governance preferences evolving over timestamp and region. The Deep Belief Network-enhanced Fuzzy Consensus Evaluation (DBN-FCE) consolidates expert weight consistency. The Social Simulation-driven Fuzzy Governance Metrics (SS-FGM) run simulation scenarios for policy changes. Finally, the Multilevel Hierarchical Bayesian Aggregation for Fuzzy Outputs (MHBA-FO) provides consistent aggregation of perspectives across district, state, and national levels. This integrated approach also reinforces the interpretability and validity of governance ranking models in the process. This is done with the added attribute of adaptability and scalability for application in real-world situations. Furthermore, the suggested system provides strategic toolkits for the use of policymakers. This is done along with electoral authorities to optimize governance-related issues in process. Along with enhanced voter engagement, and assure data-centric inclusivity into democratic processes

Enhancing Gene Selection in Microarray Dataset Using Binary Gray Wolf Optimization Algorithm and Statistical Dependence Technique

Hanadi Dawood Saleem — 2026-02-01

The problem of selecting the most relevant subset of features is vital for enhancing classification accuracy while minimizing computational load in machine learning. To tackle this challenge, the paper investigates two approaches: the Binary Gray Wolf Optimization (BGWO) algorithm and the Statistical Dependence (SD) technique. The process begins with the SD technique to determine the features that most significantly influence classification outcomes. Then, the BGWO algorithm is applied in conjunction with the K-Nearest Neighbors (KNN) classifier to further narrow down the selection to the most essential features. The proposed SD-BGWO approach outperforms conventional methods by either improving classification accuracy or by reducing the number of features required, thereby optimizing the feature selection process in terms of both efficiency and effectiveness.

Cross-Attention Feature Fusion for Interpretable Zero-Day Malware Detection

Njood aljarrah — 2026-01-03

The exponential proliferation of sophisticated zero-day malware variants poses critical challenges to traditional signature-based detection systems, necessitating advanced machine learning approaches that combine high-performance classification with transparent decision-making processes. While existing deep learning models achieve remarkable accuracy in malware detection, their black-box nature severely limits adoption in critical cybersecurity applications where interpretability is paramount for threat analysis and incident response. This work presents a novel cross-attention feature fusion architecture integrated with comprehensive explainable artificial intelligence (XAI) techniques for zero-day malware classification and attribution analysis. Our approach employs semantic feature grouping to organize heterogeneous malware characteristics into complementary structural and content-based representations, processed through specialized encoders and fused via multi-head cross-attention mechanisms that enable sophisticated bidirectional information exchange between feature groups. The integrated XAI framework combines Integrated Gradients, SHAP, and LIME techniques to provide both global and local interpretations of classification decisions. Extensive evaluation on large-scale datasets demonstrates exceptional performance: 99.97% accuracy with 0.9999 AUC-ROC on EMBER 2018 (800K samples) and 99.99% accuracy with perfect AUC-ROC on CIC-MalMem-2022 (58.6K samples). Rigorous zero-day evaluation using family-based splitting reveals robust generalization capabilities with minimal performance degradation (0.12% for EMBER 2018, 0.08% for CIC-MalMem-2022) when encountering completely unseen malware families. Ablation studies confirm the critical contribution of cross-attention mechanisms (+0.0277 AUC improvement), while XAI analysis demonstrates high consistency across explanation methods (correlation $>$ 0.84) and provides actionable insights for security analysts. Our approach uniquely combines state-of-the-art detection performance with comprehensive explainability, advancing interpretable cybersecurity AI systems and enabling transparent threat attribution analysis essential for real-world deployment.

A Bivariate Exponential Distribution with q-Exponential Marginals and its Applications

Princy T — 2026-02-01

Bivariate Gumbel’s exponential distribution is one of the most popular continuous bivariate distributions. Comprehensive studies have been done on bivariate Gumbel’s exponential model during the past few decades. In this paper, we have derived a generalized version of bivariate Gumbel’s exponential model through entropy optimization and we call this model as q-bivariate Gumbel’s exponential model. One of the major properties of the q-bivariate Gumbel’s exponential model is that its marginal densities are q-exponential distributions. Its survival function, distribution function and density function can be expressed in terms of q-exponential function, which is the q-analogue of exponential function which posses several applications in various fields. Different properties and a characterisation theorem of this distribution have been discussed. For illustrating the use of the proposed model the unknown parameters are estimated using the method of maximum likelihood estimation. A likelihood ratio test is carried out to test the goodness of fit of q-bivariate Gumbel’s exponential distribution to verify its compatibility with the existing bivariate Gumbel’s exponential model. In order to interpret the practical applicability of q-bivariate Gumbel’s exponential model a simulation study and a real data application have been carried out. From this study, we can conclude that q-bivariate Gumbel’s exponential model shows a better fit than bivariate Gumbel’s exponential model.

AI-Driven Microservice Identification for Business Process Digital Transformation: A Multi-Dependency Collaborative Clustering Approach

Mohamed DAOUD — 2025-10-10

In the era of digital transformation, enterprises are increasingly seeking to modernize their legacy monolithic systems in favor of more agile and modular architectures. Microservices have emerged as a compelling solution, offering scalability, maintainability, and independent deployment. However, the automated extraction of microservices from legacy systems—particularly when documentation is sparse or outdated—remains a complex and unresolved challenge.

This paper introduces an AI-powered, multi-view methodology for microservice identification based on business process (BP) models. Unlike traditional approaches that rely on static code analysis or single-view aggregation, our method simultaneously captures and analyzes three critical types of dependencies within business processes: control flow, data exchange, and semantic similarity. Each dependency is modeled separately and processed through a collaborative clustering framework, where AI agents exchange signals to achieve consensus-based service decomposition.

Artificial Intelligence plays a dual role in our system: it is used for semantic enrichment—via Natural Language Processing (NLP) and Sentence-BERT embeddings—and for optimizing the clustering strategy through dependency alignment and explainability metrics. A real-world case study on the Bicing bike-sharing system in Barcelona, composed of over 50 business activities, demonstrates the effectiveness and scalability of our approach. Experimental results show that the AI-enhanced model achieves superior clustering performance in terms of cohesion, coupling, and interpretability compared to baseline methods.

By integrating AI-driven analytics with business process understanding, our approach provides a robust pathway toward automated, explainable, and domain-aligned microservice extraction—supporting sustainable digital transformation at scale.

Une étude de cas réelle sur le système de partage de vélos Bicing à Barcelone, composé de plus de 50 activités commerciales, démontre l'efficacité et l'évolutivité de notre approche. Les résultats expérimentaux montrent que le modèle amélioré par l'IA atteint des performances de regroupement supérieures en termes de cohésion, de couplage et d'interprétabilité par rapport aux méthodes de base.

Linear Algebra-Based Solution of Trinomial Markov Chain-Random Walk Between an Absorbing and an Elastic Barrier

Ahmed Elshehawey — 2025-09-06

Two trinomial Markov chain-random walk (MC-RW) problems involving nonnegative integers amidst an elastic and an absorbing barrier are considered. The first has an elastic barrier at the origin and an absorber barrier at the end-state N, while the second is the opposite. Employing an unconventional approach based on eigenvalues and eigenvectors, we derive explicit formulas for the probabilities of absorption, segregation, and annihilation at the barriers. We also extract simple closed-form expressions for specific scenarios, including the semi-infinite lattice segment case.

Bayesian Conditional Autoregressive for Rainfall Modeling in East Java

Suci Astutik — 2026-01-27

Rainfall in East Java has high spatial variation, requiring a modeling approach that can capture inter-regional dependencies. This study aims to estimate rainfall patterns using Bayesian Conditional Autoregressive (BCAR) models that incorporate spatial effects, specifically the Intrinsic Conditional Autoregressive (ICAR) and Leroux CAR specifications. Parameter estimation was conducted using Markov Chain Monte Carlo (MCMC) methods to ensure convergence and posterior stability. Monthly rainfall data from East Java during the 2022–2023 were analyzed by dividing the period into the transition to the rainy season (September–November) and the rainy season (December–February). The results indicate that during the rainy season, most climatic variables, including temperature, humidity, wind direstion, and cloud cover, do not show statistically significant effects on rainfall, whereas during the transition season,wind exhibits a significant positive influence. Comparative model evaluation reveals that the ICAR model provides the best predictive performance, as indicated by the lowest Root Mean Square Error (RMSE), while the Leroux CAR model demonstrates consistent estimation of spatial dependence across both periods. Simulation results further confirm that the parameter estimators are unbiased, as evidenced by the close agreement between simulated parameters and empirical data estimates. These findings demonstrate that BCAR models, particularly the ICAR specification, are effective in capturing spatial rainfall variability in East Java. This study contributes methodologically to spatial climatological analysis and provides a foundation for future research incorporating additional covariates and extended temporal coverage to enhance rainfall prediction accuracy.

Generalized weak ε-subdifferential and applications

Abdelghali Ammar — 2026-01-02

A concept of subdifferential of a vector-valued mapping is introduced, called generalized weak ε-subdifferential.We establish existence theorems and investigate their main properties, and provide illustrative examples to clarify the construction. This construction extends and unifies several existing notions of approximate subgradients in vector optimization, including the Pareto weak subdifferential. We establish some formulas of the generalized weak ε-subdifferential for the sum and the difference of two vector-valued mappings. A relationship between the generalized weak ε-subdifferential and a directional derivative is presented. We discuss the positive homogeneity of the generalized weak ε-subdifferential. As application of the calculus rules, we establish necessary and sufficient optimality conditions for a constrained vector optimization problem with the difference of two vector-valued mappings.

Fuzzy Analytical Hierarchy Process to Optimize Supply Chain Processes in the Digital Age

Dorra Kallel — 2025-12-12

The rise of e-commerce has profoundly reshaped supply chain management in the apparel industry, increasing pressure on companies to enhance responsiveness, efficiency, and service quality. This study evaluates the influence of e-commerce on key supply chain dimensions using the Fuzzy Analytical Hierarchy Process (FAHP). Six criteria are examined: efficiency, delivery, environmental impact, services, social, and economic factors. Expert judgments, collected from professionals in the apparel sector, reveal that efficiency, delivery, and service quality are the most influential criteria in the digital context. Beyond identifying key priorities, this study provides a structured decision-making framework that supports managers in addressing uncertainty inherent to digital supply chains. The findings also highlight the strategic value of integrating fuzzy MCDM tools to guide future supply chain optimization initiatives. These insights provide strategic guidance for apparel companies seeking to improve supply chain performance and adapt to the evolving demands of online commerce.

GenAI Meets Explainability: Turning Churn Predictions into Personalized Retention Strategies

Meryem Houssam — 2026-01-17

In an increasingly competitive financial landscape, retaining existing customers is widely acknowledged to be more costeffective than acquiring new ones. While artificial intelligence (AI)-based predictive models have achieved high accuracy in identifying customers at risk of churn, they often fail to provide actionable strategies for customer retention. This paper addresses this limitation by proposing a post modeling framework that translates churn predictions into business-oriented retention actions. Using supervised machine learning techniques on structured customer data—such as transactional and behavioral features—from the financial sector, we first develop a high-performance churn prediction model. We then employ explainability methods, notably SHAP (SHapley Additive exPlanations), to identify the key drivers of churn at both global and individual levels. These insights enable us to segment customers into interpretable profiles (e.g., price-sensitive, service-dissatisfied, inactive), each associated with specific churn triggers. To move beyond prediction and toward proactive intervention, we propose tailored retention strategies aligned with each segment’s churn rationale. Furthermore, we explore the integration of Generative AI (GenAI) to support the automatic generation of personalized messages and strategy suggestions, enhancing the decision-making process for financial institutions. The proposed methodology bridges the gap between churn prediction and business actionability, offering a data-driven approach to customer engagement. Our results demonstrate that such an approach not only deepens customer understanding but also significantly improves the effectiveness of targeted retention campaigns.