The number of new cancer cases is expected to increase by about 50 % over the next 20 years, and the need for chemotherapy treatments will increase accordingly. Treatments are usually provided in outpatient cancer centers, where patients with various types of cancers receive care. The treatment delivery must be carefully planned to optimize the use of limited resources, such as consultation and examination rooms, chairs and beds for the drug infusion, medical and nursing staff. The planning and scheduling of chemotherapy treatments involve different problems at different decision levels. In this work, we focus on the operational level and jointly address the interday and intraday multi-appointment scheduling problem. We determine the day and start time of the oncologist visit and drug infusion for a set of patients to be scheduled along a short-term planning horizon. We use a per-pathology policy, where the days of the week on which patients can be treated, depending on their pathology, are known. We consider different metrics that take into account the perspectives of both the cancer center and the patients. We formulate the problem as a multi-objective optimization problem, which is tackled by sequentially solving three problems in a lexicographic multi-objective fashion. The problems turn out to be computationally challenging; thus, we propose ad hoc decomposition approaches. The approaches are tested on real data from an Italian hospital and improve upon state-of-the-art solvers.

This abstract was taken from the following paper: Carello G., Passacantando M., Tanfani E. (in press), Interday and Intraday Chemotherapy Appointment Scheduling: a Patient-Centered Approach, European Journal of Operational Research, DOI: 10.1016/j.ejor.2025.12.023.

The logistics industry plays a vital role in supporting economic growth by facilitating the efficient movement of goods and services and is a key component of Malaysia’s supply chain and trade activities. However, logistics companies are highly exposed to operational risks, economic fluctuations, and financial uncertainties, particularly during challenging periods such as the COVID-19 pandemic and its aftermath from 2020 to 2024. Without systematic financial evaluation, stakeholders such as investors, management, and policymakers may face difficulties in identifying financially healthy and distressed companies. The purpose of this study is to assess the financial performance of logistics companies listed on Bursa Malaysia from 2020 to 2024 using the Zmijewski model. A total of 28 logistics companies were evaluated in this study. The findings indicate that 24 companies remained financially healthy throughout the five-year period, representing 85.71% of the sampled firms. The results suggest that the majority of logistics companies maintained sound financial conditions despite economic challenges during the study period. The financially unhealthy companies may use the financially healthy firms as benchmarks to improve their financial performance in the future. The findings of this study are significant as they assist investors in making informed investment decisions, enable company management to benchmark their financial performance for future improvement, and support policymakers in understanding the financial resilience of the logistics sector.

This study presents a Poisson-based exceedance probability model rooted in stochastic process theory and statistical rate estimation to characterize earthquake occurrence in the Philippines using a PHIVOLCS-patterned seismic dataset. Earthquakes are modeled as realizations of a stationary Poisson counting process, where magnitude-threshold exceedances correspond to rare-event arrivals, analogous to demand or failure events in queueing and reliability systems. A 50-year catalogue comprising 5,895 events with magnitudes M ≥ 4.0 was analyzed using frequency–magnitude statistics to estimate intensity parameters for thresholds M ≥ 4.0, M ≥ 5.0, M ≥ 6.0, and M ≥ 7.0. These estimated rates were embedded in closed-form Poisson exceedance functions to compute probabilities at one-, ten-, and fifty-year horizons under standard assumptions of independence and temporal stationarity. The results yielded mean annual rates of 117.9, 14.04, 1.30, and 0.18 events for increasing magnitude thresholds, respectively. The exceedance probability of at least one M≥6.0 event was 72.7% within one year and approached unity over a decade, while M≥7.0 events exhibited a 16.5% annual exceedance probability that increased to over 99% within fifty years. From an operations research standpoint, these exceedance probabilities serve as inputs to stochastic optimization, enabling risk-informed decisions in capacity planning, resource allocation, and resilience optimization under uncertainty. The proposed framework demonstrates how classical stochastic models, widely used in queueing theory and reliability analysis, can be rigorously applied to seismic data, reinforcing the role of applied probability and statistics in interdisciplinary hazard modeling and decision science.

Introduction. Student attrition and academic underperformance remain critical challenges in higher education institutions worldwide. Early identification of at-risk learners enables timely interventions that can significantly improve student outcomes. This study compares traditional multivariate statistical techniques with modern machine learning algorithms to predict student academic performance and identify learners requiring support.

Methods. We collected comprehensive data from 525 undergraduate students over four academic years (2020-2024) through administrative records, online questionnaires (87.3% response rate), and institutional databases using stratified random sampling. The dataset comprised 18 predictor variables including prior academic achievement, behavioral factors (attendance rate, weekly study hours), socio-demographic characteristics, and engagement indicators. We systematically compared six predictive models: Multiple Linear Regression, Classification and Regression Trees, Random Forests, Gradient Boosting Machines, XGBoost, and Support Vector Regression. Model performance was evaluated using 5-fold cross-validation with R² and RMSE metrics.

Results. Feature importance analysis identified five dominant predictors: previous academic performance (34.2%), class attendance rate (21.8%), weekly study hours (15.6%), entrance examination scores (12.4%), and parental education level (8.9%). Ensemble machine learning methods substantially outperformed traditional approaches. XGBoost achieved the highest performance (R²=0.761, RMSE=0.378), representing 12-17% improvement over Multiple Linear Regression (R²=0.634, RMSE=0.487). The optimal model identified at-risk students with 84.3% accuracy, segmenting learners into four risk categories: high-achievers (23%), on-track (42%), at-risk (28%), and critical-risk (7%).

Conclusions. This research demonstrates that advanced machine learning algorithms significantly outperform traditional statistical methods in predicting student academic performance, explaining over 76% of variance. A web-based dashboard implementing the XGBoost model enables real-time risk assessment and automated alerts for academic advisors. These findings illustrate the transformative potential of data-driven approaches for evidence-based early intervention strategies and educational planning initiatives.

This work introduces a nonparametric estimator for evaluating the performability of semi-Markov systems, formulated as the sum or integral of a real-valued functional stochastic process. The concept of performability, originally proposed by Meyer, extends classical reliability measures by incorporating performance-related aspects of system behavior. It represents a unified and comprehensive measure that captures both the reliability and performance of a stochastic system as it evolves over time under uncertainty.

For a homogeneous continuous-time semi-Markov process with a given state space and reward rate function, we develop empirical nonparametric estimators for key quantities such as the semi-Markov kernel, renewal matrix, semi-Markov transition matrix, and the mean performance of the system. These estimators are constructed without imposing restrictive parametric assumptions on the sojourn-time distributions, thereby offering improved flexibility, robustness, and adaptability in practical and theoretical applications.

Furthermore, the asymptotic properties of the proposed estimators are rigorously analyzed. In particular, we establish their strong consistency and asymptotic normality, providing a solid theoretical foundation for nonparametric inference in homogeneous continuous-time semi-Markov process. Finally, the usefulness and effectiveness of the theoretical results are demonstrated through a numerical example, confirming the practical relevance and applicability of the proposed approach to performability analysis in complex stochastic systems.

Being able to make reliable predictions is a crucial task in many Artificial Intelligence problems.
In mathematical terms, a prediction can be formulated as a classification problem, where any input data is associated with a class, or as a regression task, where we search for a suitable function that fits the data.
These two classes of prediction models serve different purposes and are useful in their own regard: classification is applied to medical diagnosis, credit rating, and test grading, while regression is used to predict blood pressure, house prices, or energy consumption.
Current metrics, such as MSE for regression and AUC (Area Under the Curve) for classification, appear largely unrelated and problem-specific.
We propose a novel family of metrics, grounded in Cramér and energy distances applicable to all types of response variables: continuous, ordinal, nominal, and extendable to multivariate settings. We show the effectiveness and versatility of our metrics in a range of real applications, from finance, health care, to human resource management. We also show that metrics can be extended to assess not only accuracy but also explainability and robustness, in a joint AI assessment framework. To derive an integrated metric, we consider alternative aggregation schemes: from weighted means to decision theoretic methods.

Optimal control theory provides a powerful mathematical framework for formulating effective ecological management strategies, particularly for complex systems such as predator-prey interactions. While foundational models like the Lotka-Volterra equations offer important insights into species relationships, their use in sustainable management is limited by simplifying assumptions, particularly the omission of intraspecific competition. Modern research has addressed these limitations by integrating more realistic ecological mechanisms and applying optimal control theory as a powerful mathematical method. This study extends that approach by applying optimal control theory on a predator-prey model that includes internal competition in both species, investigating two distinct management goals. The first strategy seeks to conserve predators through prey addition, where controlled prey supplementation supports predator growth in finite time while maintaining prey availability and overall ecosystem stability. The objective is to increase the predator population while quadratically penalizing control effort through a running cost functional. The second strategy aims to maximize the total population by controlling the interaction rate between species. For this approach, a bang–bang control problem is formulated in which a linear control variable governs the mixing (or segregation) of populations, with the goal of maximizing the total population at a fixed final time. Pontryagin’s Maximum Principle is applied to derive necessary optimality conditions for both problems. Numerical simulations show that the first optimal control strategy increases predator population density in finite time through sustained prey supplementation while maintaining sufficient prey availability. For the second strategy, numerical optimization identifies a specific switching time that achieves the global maximum for the total population at the terminal time. These findings provide a theoretical foundation applicable to wildlife conservation and fishery management.

Introduction: Deadbeat controllers achieve finite-time convergence in discrete-time linear systems by driving all state variables to the origin within exactly n steps, where n denotes the system order. Despite this theoretical optimality, pure deadbeat control suffers from prohibitively large initial control magnitudes, frequently inducing actuator saturation and rendering the controller impractical for physical implementation. This fundamental trade-off between convergence speed and control feasibility constitutes a critical open challenge in digital control design.

Methods: This paper proposes a two-phase hybrid control architecture that rigorously reconciles finite-time convergence with actuator constraints. During Phase I (1 to m), an Ackermann pole-placement law repositions the closed-loop eigenvalues to prescribed locations within the open unit disk, moderating transient control amplitudes while ensuring asymptotic state reduction. At step m+1, Phase II activates a standard deadbeat formulation that eliminates the residual error in exactly n additional steps. A unified state-space formulation is derived, and Lyapunov-based stability arguments confirm that the switching does not introduce instability. The total convergence time is analytically determined to be T_c = m + n steps.

Results: Comparative simulations on a second-order discrete-time plant (n=2) demonstrate that the hybrid scheme significantly reduces peak control effort. Specifically, the peak effort drops from 9.50 in pure deadbeat to 3.80 and 3.72 for hybrid modes m=2 and m=4, respectively, achieving up to a 61% reduction. Meanwhile, the Mean Square Error (MSE) shifts from 4.13 (pure deadbeat) to 5.58 (m=2) and 5.45 (m=4). This confirms that increasing the switching instant m effectively reduces peak control magnitude at the cost of a proportional increase in settling time and transient error.

Conclusions: The proposed m-step Ackermann / (m+1)-deadbeat hybrid controller provides a formally grounded, tunable framework for discrete-time control design. By parameterizing the switching instant m, practitioners can explicitly trade off convergence speed for reduced control effort, enabling actuator-aware implementation without sacrificing the zero-steady-state-error guarantee of deadbeat control.

The Brachistochrone is the curve that provides the fastest descent of a particle sliding without friction under a uniform gravitational field. Among all curves joining two fixed points, it minimizes the travel time and represents a fundamental problem in the calculus of variations and optimal control theory, with important applications in physics, engineering and optimization.

In this paper, we present explicit parametric representations of the Brachistochrone problem without initial velocity and perform a comparative analysis for several physical and mathematical configurations. Previous theoretical studies established the existence of optimal trajectories using dynamic programming methods, but without providing explicit expressions of the corresponding curves. Such representations are essential for both practical computations and theoretical investigations, since they allow accurate evaluation of physical quantities including distance, velocity, acceleration, curvature and travel time, both in the presence and absence of gravity.

The main objective of this work is to construct and compare parametric forms of the Brachistochrone curve without initial velocity and to analyze their geometric behavior, regularity properties, and physical interpretation. Particular attention is given to the influence of model parameters on the shape of the optimal trajectory and on the associated motion characteristics. The obtained results provide new insights into the structure of the problem and contribute to a better understanding of optimal trajectory design in gravitational environments.

Classical stochastic control theory predominantly relies on Gaussian noise models driven by continuous Wiener processes. However, these models fail to capture the heavy-tailed, discontinuous jump phenomena frequently observed in modern complex systems, such as power grids under fault conditions or networks experiencing sudden cyber-anomalies. This paper investigates the finite-time stabilization and approximate controllability of a class of semilinear impulsive stochastic evolution equations defined on a separable Hilbert space and subjected to non-Gaussian α-stable Lévy noise.

When the perturbation paradigm shifts from continuous diffusion to discontinuous jump processes, standard feedback control strategies designed under strict second-moment assumptions often experience severe performance degradation or total loss of stability. Because the variance of the Lévy process diverges for α∈(1,2), classical mean-square L² stability analysis entirely collapses. To address this fundamental limitation, we shift our analytical framework to the Banach space L^p(Ω, H) of fractional-moment integrable processes, where the moments remain finite for 1<p<α.

Within this space, we establish finite-approximate controllability by formulating a parameterized feedback control law and constructing a nonlinear operator mapping based on the mild solution of the impulsive system. By employing Picard iterations and the Banach fixed-point principle under the p-th moment norm, we theoretically guarantee the existence of a unique stabilizing state trajectory that compensates for both continuous nonlinear drift and discrete impulsive shocks.

To complement the theoretical framework, we present a preliminary numerical investigation to assess controller degradation explicitly. Through extensive simulations utilizing the Chambers–Mallows–Stuck method to generate symmetric α-stable increments, we demonstrate the failure modes of classical controllers under heavy-tailed perturbations. Ultimately, this work illustrates how control adjustments based on fractional moments facilitate finite-time stabilization, establishing a mathematically rigorous foundation for the synthesis of resilient controllers in non-Gaussian, impulsive stochastic environments.