Post-baccalaureate guidance has a lasting impact on learners’ trajectories. To address this challenge, the multinomial logistic regression model developed by previous authors recommends a post-baccalaureate track based on academic grades and transversal competencies, with high accuracy. Despite this performance, the RLM has three structural limitations: socio-contextual blindness, the absence of predictive uncertainty quantification, and temporal rigidity. To overcome them, this article proposes a Contextual Bayesian Model (MBC) that enriches the input vector with five socio-educational determinants (family income, parents’ socio-professional category and educational attainment, geographic area, digital access), treats the parameters as Gaussian random variables conditioned on context, introduces a confidence index based on Shannon entropy, and enables sequential -learning updates. Evaluated under conditions identical to those of the multinomial logistic regression model (RLM), on the same synthetic dataset, the MBC markedly improves accuracy and stability, eliminates cases of high predictive uncertainty, and its superiority is confirmed by a paired t-test. A first empirical validation on 209 candidates admitted to ESATIC, restricted to a binary SM/SP choice, confirms its operation beyond the synthetic setting. To our knowledge, it constitutes the first contextual Bayesian extension of an AI-based post-baccalaureate guidance system in the Ivorian context.
The interval-valued analysis has several inherent drawbacks, including the use of the uniform distribution over the interval, neglecting the degree of indeterminacy when data are recorded under uncertainty, and its convergence to the special case of classical statistics. Neutrosophic analysis, also known as generalized interval statistics, extends the framework of interval-valued analysis and is employed in this study for constructing the control chart. The degree of indeterminacy for the interval-valued data is calculated, and the uniform, beta, and log-logistic distributions are fitted to this degree, with the mean of these distributions used in the plotting statistic. From the analysis, it is evident that the uniform distribution does not fit well to the degree of indeterminacy. It is also clear from the analysis that simulation using interval-valued analysis cannot be recommended since in this case, the control chart results do not converge to those of the Shewhart control chart when uncertainty is absent. We also find that simulations based on the uniform distribution do not fully capture the uncertainty in the recorded interval-valued data and provide incomplete information about the process. From the analysis based on the average run length and power of the control chart, it is again evident that both the uniform distribution and interval-valued analysis cannot be recommended due to several drawbacks. Based on the findings, the neutrosophic analysis, which is a generalization and a more informative and flexible alternative to the interval-valued analysis, is recommended for control charting using interval-valued data.
Our aim is to classify gender using machine learning by analyzing user behavior on the Internet. With approximately 65% of the global population online, understanding users their preferences has become necessary. Online experiences are shaped by behavioral analysis, which enables platforms to tailor content by highlighting user interests and filtering out irrelevant information. Gender identification plays a significant role in user profiling for optimizing search results, recommending products or services, detecting fake connections, and improving social network suggestions. This study investigates user feedback on information-searching preferences and practices, treating them as indicators of online behavior, and trains a Ridge classifier to predict gender. The model achieves 70% accuracy on test data, demonstrating measurable differences in how males and females seek and disseminate information online.
High-stakes standardized exams often permit retesting, requiring admissions systems to consolidate multiple attempt scores into a single reported metric. This paper examines the distributional and policy implications of three common aggregation rules: reporting the last attempt, the average of all attempts, or the maximum (best) attempt. We develop a unified framework combining a capped zero-truncated Poisson model for heterogeneous retake behavior and a multivariate normal process with AR(1) correlation to capture within-student score dependence. Analytical results yield closed-form moments for the capped retake-count model and an exact conditional distribution for the average-score rule, clarifying when averaging improves reliability. For the maximum-score rule, we formalize mechanical upper-tail
This study examines the behavior of Nigeria's quasi-money, focusing on its nonlinear dynamics and volatility from January 1994 to December 2024. By utilizing nonlinear GARCH-type models, specifically the smooth transition autoregressive-GARCH models (LSTAR-GARCH and ESTAR-GARCH), the study indicates notable conditional heteroskedasticity, volatility clustering, and nonlinearity in the series of monetary aggregates. Diagnostic tests such as Chow and BDS highlight the existence of structural breaks and behavior dependent on different regimes, emphasizing the drawbacks of conventional linear models. Among the various models, the LSTAR-GARCH shows better statistical results and is more reactive to abrupt changes in economic or policy conditions. These results stress the essential role of using nonlinear and regime-switching volatility models to analyze quasi-money in developing countries, equipping policymakers with improved methods for forecasting and handling liquidity during structural shifts. The results indicate that those in charge of monetary policy should use nonlinear and regime-switching volatility models, especially the LSTAR-GARCH model, in their analysis to enhance effective prediction of the changes in Nigeria's quasi-money. Moreover, it is advised to frequently check for structural breaks for timely policy updates. Financial analysts, too, should focus on models that can capture both nonlinearity and changing volatility to achieve better forecasts and a clearer understanding of monetary trends.
We study the method for detecting differentially expressed genes (DEGs) in two groups using RNA-seq data. The primary method employed was the test-based methods, although recently, clustering-based methods have been proposed. But, the sample size of RNA-seq data is too small to be tested in many cases. Moreover, the number of true DEGs affects the accuracy of clustering-based methods. Therefore, we propose a method for transforming expression levels to detect DEGs. Because the expression levels after transformation showed different expression patterns only for DEGs, the isolation forests were used to detect them. We compare this method with conventional methods using simulations. When the total number of DEGs was small, the proposed method was the most accurate for any simulated situation. Simultaneously, the proposed method was effective only when the expression level of the experimental group was higher than that of the control group. However, if the expression level transformation equation was changed and the proposed method reapplied, genes with higher expression levels in the control group could be detected. The proposed method was more likely to detect genes with a large fold-change than the conventional method.
The widespread improvement of digital financial literacy has emerged as a critical challenge for inclusive development within the global digital economy. Whether and how large-scale digital infrastructure construction can contribute to this process remains a key concern for both policymakers and academic researchers. Using the Broadband China demonstration city policy as a quasi-natural experiment, this study draws on data from the China Household Finance Survey (2011-2019) and employs a multi-period difference-in-differences approach to systematically examine the causal impact of digital infrastructure on residents’ digital financial literacy and its underlying mechanisms. The results show that the policy significantly enhances residents’ digital financial literacy, an effect that remains robust across a series of tests. Heterogeneity analysis reveals structural variations in policy effects, regionally, the impact is stronger in eastern China than in the central and western regions; at the individual level, it is more pronounced among low-education and unmarried groups, reflecting the policy’s inclusive and empowering nature. Further mechanism analysis indicates that smartphone use and digital payment adoption serve as key transmission channels, with high-frequency digital payment practices playing a particularly significant role. By integrating macro-level policy, micro-level behavior, and individual capability into a coherent framework, this study provides empirical evidence from a large-scale natural experiment for understanding how financial literacy is formed in the digital age. It also offers relevant policy insights for countries - especially developing economies - seeking to advance financial inclusion and human capital accumulation through public digital investment.
Purpose: By expanding the Theory of Planned Behavior (TPB), this study explores what drives Egyptian consumers to buy health insurance. The study assesses how insurance literacy, attitude, perceived behavioral control, and subjective norms shape purchase intentions and subsequent buying behavior. Furthermore, it analyzes whether intention acts as a mediator in these relationships and if trust and gender function as significant moderators. Design/methodology/approach: This study adopted a quantitative approach, gathering primary data through an administrative survey of 290 Egyptian insurance policyholders. To validate the proposed conceptual framework, the research utilized Partial Least Squares Structural Equation Modeling (PLS-SEM), executed through the SmartPLS 4 software suite. Findings: Based on the data analysis, insurance literacy emerged as the primary predictor of purchase intention among the variables studied. Positive attitudes and perceived behavioral control also demonstrated significant positive effects on intent. However, the influence of subjective norms was found to be statistically insignificant, suggesting that social pressure plays a minimal role in the Egyptian health insurance context. Moreover, intentions to purchase positively influence actual purchase behavior. The analysis confirms that purchase intention serves as a vital mediator, successfully translating cognitive and educational antecedents into actual behavior. While gender was found to significantly moderate the transition from intention to action indicating distinct behavioral pathways for men and women, customer trust did not emerge as a significant moderator. Practical implications: For policymakers and insurance providers, the results suggest that educational campaigns aimed at increasing insurance literacy are more effective for driving enrollment than trust-based or social-influence marketing. Furthermore, the significant gender moderation implies that practitioners should adopt segmented strategies that address the unique barriers and drivers relevant to different gender cohorts. Originality/value: The study contributes to the literature by including insurance literacy into the TPB framework within the context of a developing economy undergoing healthcare reform. It provides rare empirical evidence on the “intention-behavior gap” and clarifies how structural and demographic variables influence the acquisition of health coverage in a transitioning market.
Barometric pressure plays a pivotal role in atmospheric circulation and synoptic-scale weather dynamics. This study develops a three-state discrete-time homogeneous Markov chain framework to model and forecast barometric pressure regimes, analysing data recorded by the central pollution control board (CPCB), PUSA, Delhi, over the period 2018-2024. The continuous pressure series was categorized into Low, Normal and High regimes using global percentile-based thresholds $(P_33 and P_67)$ to ensure balanced and physically interpretable state representation. The stochastic framework was applied at daily, monthly, seasonal and annual temporal scales, covering regime persistence, transition characteristics, probabilistic structure, and predictive performance. Estimated transition probability matrices reveal scale-dependent behaviour, with strong self-transition probabilities at daily resolution indicating substantial short-term persistence in atmospheric pressure dynamics. Stationary distributions and mean recurrence times confirm the probabilistic stability of the three-state structure. Model adequacy was evaluated using chi-square goodness-of-fit statistics; both AIC and BIC values support the first-order Markov assumption. The Diebold-Mariano test was used to compare forecast accuracy across the Markov, auto regressive integrated moving average (ARIMA), and linear regression models, with MAE, RMSE and MAPE as performance metrics. At the daily scale, the Markov model outperformed both ARIMA and linear regression, though performance differences narrowed at monthly and seasonal resolutions. The results confirm that multi-scale regime-based stochastic modelling provides an interpretable and computationally efficient framework for analysing and forecasting barometric pressure. Pressure regime dynamics, including atmospheric persistence and scale-dependent variability, were better captured by transition probability models than by conventional alternatives.
This article examines recent advances in the assessment and assurance of e-learning quality in higher education (HE) through a comprehensive review supported by a large-scale bibliometric analysis of 34,325 publications (2015-2025). The study identifies major scientific trends, including the expansion of blended and digital learning models, and the growing emphasis on learner engagement and institutional performance. Using lexical analysis and topic modeling, the review highlights the multidimensional nature of e-learning quality, spanning pedagogical, technological, organizational, and human factors. Established and emerging frameworks are analyzed alongside international standards. The findings reveal persistent challenges, including the lack of universal standards and the gap between technological innovation and quality assurance practices. The article concludes by advocating for hybrid, adaptive, and learner-centered quality models capable of supporting sustainable digital transformation in higher education. Received: December 13, 2025Revised: March 7, 2026Accepted: March 21, 2026
A major challenge in analyzing time-dependent data, like in survival analysis, occurs when the precise timing of events is not observed but is only known to fall within a certain time interval, a situation known as interval censoring. This study investigates the impact of interval censoring on the performance of the Cox proportional hazards model, combining both simulation experiments and empirical evidence. In the simulations, four event probability scenarios $(p=0.05, 0.125, 0.25, 0.5)$ were generated with 1,000 cases each; a scenario with 500 and 5,000 cases was also applied. Event times were drawn from a uniform distribution $[0, 1]$, analyzed under continuous observation and interval-censored frameworks with 2, 4, 5, and 8 subintervals. Model performance was evaluated using the likelihood ratio (LR) test. Findings revealed a substantial loss of accuracy under interval censoring, particularly in higher event probability settings, though increasing the number of intervals progressively improved the fit, with eight intervals approaching the continuous-time estimates. To complement the simulations, empirical data were analyzed under similar scenarios. Results confirmed that interval censoring significantly affected model performance. In conclusion, interval censoring compromises the reliability of Cox models, emphasizing the importance of frequent follow-up assessments or the application of specialized methods for interval-censored data to obtain valid and robust survival estimates.
The production of palay (unmilled rice) is important for ensuring food security and sustaining livelihoods in the Philippines, with palay serving as a major contributor to agricultural output. In Davao Oriental, while palay production remains significant, studies focusing on the behavior of land utilization for palay cultivation are limited. This study aims to model and forecast the quarterly area harvested for palay in Davao Oriental using a Seasonal Autoregressive Integrated Moving Average (SARIMA) approach. Results show seasonality in the series, which was addressed through first seasonal differencing. The selected SARIMA (0, 0, 1) (0, 1, 0) [4] model demonstrated good forecasting performance, with statistically significant parameters and well-behaved residuals. Forecasts for the period from the fourth quarter of 2025 to the fourth quarter of 2028 indicate that the harvested area for palay is expected to remain stable, exhibiting seasonal fluctuations without a clear upward or downward trend. These results indicate the applicability of SARIMA models in agricultural land-use analysis and offer practical information for regional agricultural planning and policy formulation.
This study develops a novel framework for assessing the reliability of stress-strength models when both stress and strength follow independent inverse exponential distributions. Likelihood-based procedures are proposed for constructing confidence intervals for the reliability measure, and their performance is rigorously evaluated through extensive simulation studies across a range of parameter configurations. The results demonstrate the effectiveness of the proposed methods in characterizing lifetime behavior modeled by the inverse exponential distribution. The findings contribute to applied statistics with practical relevance in areas such as engineering reliability, medical survival analysis, and related fields.
On highways, congestion is caused not only by traffic volume but also by manual toll collection. To accelerate processing and reduce delays, electronic tolling systems have been implemented for more than two decades, yet their operational efficiency and technological reliability remain under evaluation. This study applies multinomial logistic regression models to optimize automatic vehicle classification. Two complete vehicle-passage samples were used: one to train the model and another to test its accuracy. The proposed approach achieved over 99.9% correct classification in both samples, with errors limited to vehicles whose dimensions lie on the borderline between predefined classes. These findings demonstrate that statistical modeling and classification can improve the performance and reliability of electronic toll systems, providing a robust framework for transport infrastructure management.
A semiparametric mixed switching meta-regression model with fixed covariates (possibly high dimensional) and a random component resulting from a regime switch is postulated. The model is estimated through a hybrid of cubic smoothing splines and restricted maximum likelihood estimation embedded in the backfitting algorithm. The
Paddy production in Malaysia faces significant fluctuations, maintaining a self-sufficiency ratio of approximately 65% with the remainder imported from neighbouring countries. Traditional time series models often fail to capture the complex and uncertain nature of paddy production data. This study proposes a hybrid model integrating fuzzy temporal preprocessing with LSTM to forecast paddy production across six Malaysian states. Fuzzy logic handles climate uncertainty via triangular membership functions while LSTM captures long-term temporal dependencies, combined through an equal weight linear ensemble. Model performance is evaluated by comparing the proposed model with LSTM and fuzzy WM. The results show that the proposed model achieves higher accuracy in learning vague patterns and long-term dependencies. This prediction model offers valuable insights for policymakers by improving paddy production projections and enhancing food security in Malaysia.
This study investigates the relationship between cumulative grade point average (CGPA) and the acceptance ratio of students in the Faculty of Science at the University of Tabuk, with particular emphasis on variations across levels of academic risk. Academic underachievement remains a major challenge in higher education, as traditional approaches based on average performance often fail to capture variability among academically vulnerable students. Addressing this limitation is essential for developing more effective and targeted academic support strategies. Using administrative academic records from the Faculty of Science, the study applies quantile regression analysis to assess how the effect of CGPA varies across acceptance ratios. The results indicate a consistently positive and statistically significant relationship across the lower, median, and upper quantiles, though the strength of the association differs across performance levels. The strongest effect is observed at the median quantile, providing evidence of distributional heterogeneity. These findings suggest that improvements in CGPA are most impactful for students with moderate academic difficulty, while those at higher risk may require additional, targeted support. Accordingly, the study recommends the adoption of quantile-based academic monitoring systems, risk-specific interventions, strengthened early warning mechanisms, and the integration of distribution-sensitive analytical methods into institutional decision-making. Overall, the study confirms that quantile regression provides a robust foundation for evidence-based academic policies and more effective support strategies for underachieving students at the University of Tabuk.
The cryptocurrency market has high volatility and a dynamic nature, which poses significant challenges for long-term predictions and investment decision-making. This study investigates approaches that sentiment analysis. The evaluation process includes testing of and bi-directional LSTM. Logistic regression achieved an accuracy of 87% while VADER outperformed all other models with an accuracy of 93%. The XGBoost regressor achieved better Bitcoin price forecasting results when technical indicators and on-chain data were value of 2,031.56. The model achieved better results after sentiment data integration because it produced an RMSE value of 1,926.24. This decision-making.
This article investigates the forecasting performance of econometric and hybrid models for the consumer price index (CPI). Given the structural persistence, seasonality, and nonlinear dynamics characteristic of inflation in emerging economies, a two-stage hybrid modeling framework is employed. In the first stage, a SARIMAX model incorporating CPI sub-indices as exogenous variables captures the linear and seasonal components of inflation. In the second stage, the residual nonlinear dynamics are modeled using GARCH, Random Forest, XGBoost, LSTM, and Seq2Seq architectures. The forecasting performance of the hybrid SARIMAX-ML and SARIMAX-DL models is evaluated using a sliding-window approach combined with bagging, enabling multi-horizon predictions over a 12-month horizon. The results indicate that the SARIMAX model provides a strong baseline, while the SARIMAX-LSTM hybrid model achieves the lowest forecast errors. Overall, the findings suggest that CPI dynamics are primarily driven by stable linear relationships and seasonality, complemented by residual nonlinear structures effectively captured by recurrent neural networks. These results underscore the relevance of hybrid modeling strategies for macroeconomic inflation forecasting in emerging economies.
This research offers an actuarial and econometric examination of life insurance policy surrenders within the Egyptian market. The study utilizes quantile regression to analyse the differential effects of policy attributes at various surrender value levels, in addition to employing cluster analysis to categorize policyholder behavior. Analysis of a sample of 3,000 actual surrender instances utilizing $R$-language indicates that the influence of the sum insured on surrender value is inconsistent; it markedly increases from 0.35 at the lower quantile to 0.74 at the 90th quantile. This signifies a “pro-wealthy” bias in the existing actuarial surrender tables. Additionally, cluster analysis revealed three distinct risk categories, with “High-Net-Worth Investors” presenting the most substantial liquidity risk. The report advocates for the reconfiguration of surrender value schedules to guarantee equitable distribution and the establishment of early-warning systems for managing liquidity risk.
