Abstract
Handling imbalanced target distributions in regression poses a persistent challenge, as the underrepresentation of relevant target values can significantly hinder model performance. Existing data-level solutions often adapt classification-oriented techniques, introducing arbitrary thresholds over the continuous target and leading to artificial and potentially misleading problem formulations. Deep generative models offer flexible sample synthesis but are computationally intensive and difficult to interpret. We propose a CART-based synthetic sampling method specifically designed for imbalanced regression on tabular data. The method integrates relevance- and density-guided sampling to address sparse target regions without thresholding, and employs a feature-driven tree structure to generate realistic tabular samples across heterogeneous features and non-linear interactions. Experiments on benchmark datasets for extreme-value prediction show that the proposed approach is competitive with state-of-the-art resampling and generative methods while offering faster execution and greater transparency. These results highlight its potential as a scalable and interpretable data-level strategy for improving regression models in imbalanced domains. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.

Abstract
Forest management and operations planning involve complex decisions that integrate ecological knowledge, spatial data, and analytical tools to balance sustainable resource use with risk mitigation. Disturbances such as storms, diseases, and wildfires increasingly disrupt forest ecosystems and value chains. The timely removal, processing, and delivery of forest residues to bioenergy facilities are essential to reduce wildfire risk, prevent disease spread, and ensure operational continuity for forest managers and owners. This study presents a decision-support approach to address supply uncertainty caused by wildfires within the forest-to-bioenergy value chain. The methodology first generates multiple raw material variability scenarios using a fire simulation model, then clusters them according to post-fire biomass availability and probability of occurrence. These clusters are integrated into a two-stage stochastic optimization model incorporating a Conditional Value-at-Risk (CVaR) metric. Results show that the stochastic model with CVaR achieves the lowest total cost while ensuring complete processing of biomass under the most severe wildfire scenarios. The findings highlight the value of flexible and risk-aware planning strategies for forest operations, supporting decision-makers in balancing investments in processing capacity, cost efficiency, and post-disturbance resource utilization.

Abstract
Growing urbanisation, the development of smart cities, and environmental concerns have driven the implementation of advanced technologies and the modernisation of transport systems. Electric motorcycles have emerged as an effective solution for mobility, but they also present specific challenges, particularly related to the mode of riding, which is more complex than that of other vehicles and requires greater attention, skill, and preparation. Therefore, the interaction between the rider and the support system must be carefully designed, with particular emphasis on the interface and the adaptation of route information. This interface should be intuitive, accessible, and capable of presenting relevant information in a clear and objective manner, minimising distractions while riding. In addition, it must be adaptable to user preferences, allowing for customisation such as colour themes, levels of detail in the information displayed, or specific notifications regarding adverse weather and road conditions. Adapting route information provides a more efficient, safe, and satisfactory user experience. It enables riders to access personalised information, continuously updated in real time and tailored to the situation and their specific needs, including traffic conditions, road surface state, and weather conditions. This optimisation leads to better time management, energy consumption, and overall ride quality, enhancing urgent and non-urgent services. Moreover, integrating clearly and objectively features such as voice commands and compatibility with mobile or wearable devices (e.g., smartwatches) can facilitate real-time interaction without compromising safety. The interface should also offer advanced functionalities in line with technological developments and user needs, adapting to each rider’s specific requirements. This not only improves the individual experience but also promotes efficiency and sustainability, contributing to the advancement of smart cities and innovative mobility solutions. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.

Abstract
[No abstract available]

Abstract
The Benefit of the Doubt (BoD) is a non-parametric frontier model derived from Data Envelopment Analysis (DEA), used to construct composite indicators in various sectors of economic activity, with a particular focus on macroeconomic assessments. Based on documents published in the Web of Science from 1991 to 2025, we conduct a systematic bibliometric review on this topic, proposing future research directions derived from the bibliographic coverage of the most recurrent concepts, areas, and problems addressed in the current BoD literature. We identify core publication networks for non-parametric frontier composite indicators, highlighting trends, hot topics, and clusters of applications. As a result, we offer three different and comprehensive BoD research agendas based on a practical knowledge discovery exercise from expert knowledge and Large Language Models (LLM), highlighting attractive topics, theoretical contributions, concepts, methods, and potential applications. © 2025 The Author(s)

Abstract
This study highlights the critical role of Large Language Model (LLM) in simplifying technical content and integrating visual data for accessible communication. It compares GPT-4 and Llama-3.2-90b-Vision-Preview, focusing on readability, semantic similarity, and multimodal interpretation using robust metrics like Flesch Reading Ease, Gunning Fog Index, and CLIP Score. GPT-4 retains key information and achieves high semantic and textual integration scores, making it more suitable for complex technical scenarios. Furthermore, LLaMA prioritizes readability and simplicity, outperforming in generating accessible captions. Both models show optimal performance with a temperature setting of 0.5, balancing simplicity and meaning preservation. The research underscores LLM potential to democratize technical knowledge across disciplines but notes precision and multimodal integration limitations. Future directions include fine-tuning for domain-specific applications and expanding input modalities to enhance accessibility and efficiency in real-world technical tasks.