This study proposes a comparative methodological framework for objective Indoor Environmental Quality (IEQ) assessment in cultural heritage (CH) buildings, integrating two complementary scoring approaches. The first approach follows a conventional threshold-based frequency method, classifying occupied-time observations as acceptable or non-acceptable according to standard-defined limits. The second approach extends this perspective by introducing a continuous distance-based formulation, where each observation is evaluated based on its normalized distance from an ideal range and transformed into a utility score. This continuous representation enables a more informative assessment of environmental conditions by capturing the magnitude of deviations from optimal levels, rather than relying solely on binary compliance. The framework is demonstrated using monitored data from a cultural heritage building in Croatia. The proposed framework supports a more comprehensive understanding of IEQ performance and facilitates improved decision-making in heritage building assessment contexts, where both occupant comfort and conservation requirements must be considered.
The phenomenon known as the curse of dimensionality persists as a significant barrier to advancement in machine learning applications within biomedicine and other domains characterized by limited data availability. The challenges posed by high-dimensional, low-sample datasets often lead to issues such as overfitting, instability, and the emergence of spurious findings. This situation creates a perceived dichotomy between achieving predictive accuracy and maintaining interpretability. In this study a systematic head-to-head comparison of LASSO, Elastic Net, AdaBoost, and LogitBoost is introduced in the context of extreme p ≫ n conditions. Boosting demonstrated superior performance relative to all competitors, attaining the highest levels of predictive accuracy alongside the stable and interpretable feature sets. Elastic Net demonstrated robust performance in scenarios characterized by correlated features, whereas boosting methods, traditionally regarded as optimizers of accuracy, exhibited greater instability and diminished reproducibility. This study illustrates that the commonly held belief in an accuracy–interpretability trade-off can not be effectively surmounted. The implications of these findings extend significantly across the fields of proteomics, and precision medicine, particularly in contexts where the costs associated with sample collection are substantial, yet the need for accuracy remains paramount.
In this study, we define the user-specific tacit elements embedded in generative-AI use as tacit knowledge and propose a system that supports organizational collective intelligence by converting and sharing them as explicit knowledge based on the SECI model. Specifically, we construct each SECI process (Ba) as a multi-agent system to improve the autonomy and sustainability of the knowledge-creation process. Regarding the proposal system, we implemented the process equivalent to Socialization as a web application, and designed and evaluated the other processes (Externalization to Internalization) as prototypes on an API platform. The pilot study verified that parts of the process from socialization to internalization were observed. This suggests that the accumulation of further case studies may solidify the utility of this system. These results indicate that this system has the initial feasibility to be an effective foundation for promoting and supporting organizational knowledge creation.
A logistics system involves the movement of goods and vehicles across various locations, aiming to minimize costs such as distance traveled, fuel consumption, and delivery time. One of the core tasks in logistics is determining the best route for a vehicle to visit a set of destinations and return to a base, which is precisely the essence of the Traveling Salesman Problem (TSP). In this article we consider four competing formulations for solving the undirected Traveling Salesman Problem (TSP), namely: mixed-integer linear programming (MILP), nearest neighbor algorithm (NNA), genetic algorithm (GA) and simulated annealing (SA). Based on the experimental results the running time of the exact MILP increases exponentially as the number of delivery locations increases. In contrast, the heuristic approaches find good approximate solutions in a fraction of the time required by the exact technique.
This paper proposes a methodological framework for explainable thermal comfort measurement in human-centric smart buildings. The proposed framework combines environmental sensing, occupant feedback, machine learning-based comfort measurement, and explainability mechanisms (XAI) within a unified decision-support pipeline. The methodology introduces a two-level explainability architecture aimed at improving the interpretability and robustness of comfort measurement. The first level gives a sensitivity-oriented interpretation by quantifying the contribution of each measured variable to the predicted comfort condition. The second level extends explainability toward actionable decision support by generating measurement-oriented comfort recommendations. Starting from the measured indoor conditions, the framework identifies the minimum feasible environmental adjustments required to shift the occupant comfort state from discomfort toward an acceptable comfort state. The paper transforms XAI from a passive interpretation tool into a metrological layer to support personalized HVAC management. The framework contributes to the development of trustworthy, and measurement-informed thermal comfort services for next-generation smart building ecosystems.
