Domenico Santoro

A network risk associated with persistent exposure depends on both the history of the exposure and the points at which decisions can be made. These two sources of dependence must remain distinguishable when an arc is divided into shorter segments. Substochastic operators provide a finite-state description in which homogeneous subdivision preserves propagation, while changes in exposure and in the available continuations remain visible. Building on continuous semigroups and linear realization theory, this paper studies the interaction between refinement and the information required for network decisions. At each physical vertex, the pairing of reachable states with admissible continuation observations determines the dimension of an exact linear state. Its transitions respect arc subdivision when the physical decision and stopping locations are preserved. The same observations determine which partial paths can be discarded without worsening any permitted continuation. Worked finite-state networks demonstrate why a smaller prefix risk may lead to a larger final risk and how recovery, exposure ordering, and generator uncertainty affect that comparison. They also show that allowing uncertainty to change at an artificial breakpoint alters the decision problem even when local propagation remains consistent. These relationships connect physical exposure, state reduction, and path optimization within a formulation whose invariance and computational requirements can be stated explicitly.

Despite significant advances in tourism forecasting methods, current approaches suffer from critical limitations including static ensemble weighting mechanisms that fail to adapt to changing environmental conditions, insufficient integration of multi-source data streams, and limited robustness against sudden demand shifts caused by extreme weather or unexpected events. This study presents an innovative ensemble artificial intelligence framework for monitoring and forecasting tourist flows in the Aosta Valley region, Italy, utilizing a large-scale dataset of over 41 million vehicle passages collected from 14 strategically positioned sensor portals. Our novel approach integrates multiple machine learning algorithms through an adaptive ensemble mechanism that dynamically weights individual predictors based on temporal patterns, seasonal variations, and real-time performance metrics. We introduce the Adaptive Temporal Ensemble (ATE) algorithm, combining eXtreme Gradient Boosting (XGBoost), Random Forest, Support Vector Regression, and Long Short-Term Memory networks with a novel meta-learning layer. The key novelty lies in the dynamic weight adjustment mechanism that responds to contextual features including recent model performance, seasonal indicators, meteorological conditions, and traffic flow characteristics, enabling the system to automatically select the most appropriate predictor for each forecasting scenario. The system processes traffic data from highway and valley road sensors, integrated with comprehensive meteorological datasets and calendar information, providing real-time monitoring and accurate forecasting capabilities. We present a formal mathematical framework, including the Ensemble Convergence Theorem, which guarantees optimal performance bounds under specific conditions. Experimental validation demonstrates superior forecasting accuracy with Mean Absolute Error (MAE) improvements of 23.7% and Mean Squared Error (MSE) reductions of 31.2% compared to individual models. The ensemble framework achieves R ² scores exceeding 0.94 for short-term predictions and maintains robustness across different seasonal patterns and extreme weather conditions. These improvements translate directly into practical benefits for destination management organizations, including enhanced resource allocation efficiency, improved traffic congestion management, and more accurate capacity planning for tourism infrastructure. This research contributes significantly to intelligent tourism management systems and provides a scalable framework applicable to other regions with similar traffic monitoring infrastructure.

Open data and open-source software are vital for fostering scientific research transparency, reproducibility, and collaboration. By leveraging open data, researchers can access high-quality information without barriers, while open-source tools provide the flexibility and customization needed for innovative solutions. In this study, we used Python to freely download urban maps of some Italian municipalities belonging to the Daunian Mountains area (province of Foggia, Apulia, Italy), represented as graphs. These maps served as the basis for addressing the Mixed Chinese Postman Problem (MCPP), a well-known NP-complete problem in graph theory. The MCPP involves finding the shortest route that traverses all the edges of a mixed graph at least once, with the challenge of considering the difference between directed and undirected edges. To solve the MCPP, we implemented an artificial intelligence approach using Ant Colony Optimization (ACO), a bio-inspired metaheuristic that simulates the behavior of ants in searching for efficient paths. Our article details the entire workflow, from data acquisition and preprocessing of urban graphs to the optimization and analysis of ACO hyperparameters. Through an extensive optimization process, we identified the most effective hyperparameter configurations for solving the MCPP on these graphs. We also assessed the strengths and limitations of metaheuristics when applied to NP-complete problems. Aligned with the principles of open science, we release the original graph datasets of the Subappennino Dauno municipalities, together with the Python codes used for graph extraction and the best solutions obtained in the experiments, making them available for reuse and testing of alternative optimization methodologies by other researchers. This work demonstrates how integrating open data and open-source tools can address complex scientific challenges and foster collaborative research. Moreover, using ACO enhances the model’s explainability by enabling the identification and analysis of hyperparameter patterns concerning the problem’s structural characteristics. This approach allows for a deeper structural understanding of the problem itself and provides insights into the theoretically optimal techniques that can be employed to address it.

Sudden Cardiac Death (SCD) is a critical and unexpected condition that occurs due to cardiac causes within one hour of the onset of acute cardiovascular symptoms or twenty-four hours in unwitnessed cases. Despite advancements in cardiovascular medicine, practical methods for predicting SCD are still lacking, and there are no standardized systems to identify individuals at risk, especially in seemingly healthy populations such as athletes. In this study, we employed hierarchical clustering and principal component analysis (PCA) on data from 711 competitive athletes, revealing distinct patterns and cluster distributions in PCA space. Specifically, Clustering revealed characteristic feature combinations associated with increased SCD risk in athletes. Notably, certain clusters shared traits, including participation in Class C sports, sinus tachycardia, ventricular pre-excitation, personal or family history of heart disease, T-wave inversions, and prolonged QTc intervals. PCA helped visualize these patterns in distinct spatial regions, highlighting underlying structures and aiding intuitive risk interpretation. These results enable scientists to derive cluster metrics that serve as reference points for classifying new individuals and visually representing risk patterns in a clear graphical format. These findings establish a foundation for predictive tools that, with additional clinical validation, could aid in the prevention of SCD. The dataset used in this study, along with the clustering and PCA results, is available to the scientific community in an open format, together with the necessary tools and scripts to enable independent experimentation and further analysis.

This study presents an advanced system for monitoring and forecasting tourist flows in the Aosta Valley using distributed sensor technologies, cameras, and machine learning algorithms. This innovative system is designed to provide real-time data on arrivals and presences throughout the region, helping to manage traffic and tourism resources more effectively. The research analyzes data collected from portals equipped for traffic detection. Through a multi-phase approach, the project integrates and analyzes over 41 million vehicle passages to support informed decisions for regional economic and social policies. Furthermore, computational processes were conducted to optimize the analysis of the vehicle flow, reducing the dataset and focusing on checkpoints and vehicle categories. This type of time series revealed high stationarity, allowing the use of the eXtreme Gradient Boosting (XGBoost) algorithm for more accurate forecasts than Deep Learning models and other Machine Learning algorithms, such as those highlighted in terms of MAE and MSE. The results represent a significant step forward in managing tourist flows and improving the Aosta Valley’s operational efficiency and visitor experience.