Abstract

Abstract
The electrification of the transport sector is a critical element in the transition to a low-emissions economy, driven by the widespread adoption of electric vehicles (EV) and the integration of renewable energy sources (RES). However, managing the increasing demand for EV charging infrastructure while meeting carbon emission reduction targets is a significant challenge for charging station operators. This work introduces a novel carbon-aware dynamic pricing framework for EV charging, formulated as a chance-constrained optimization problem to consider forecast uncertainties in RES generation, load, and grid carbon intensity. The model generates day-ahead dynamic tariffs for EV drivers with a certain elastic behavior while optimizing costs and complying with a carbon emissions budget. Different types of budgets for Scope 2 emissions (indirect emissions of purchased electricity consumed by a company) are conceptualized and demonstrate the advantages of a stochastic approach over deterministic models in managing emissions under forecast uncertainty, improving the reduction rate of emissions per feasible day of optimization by 24 %. Additionally, a surrogate machine learning model is proposed to approximate the outcomes of stochastic optimization, enabling the application of state-of-the-art explainability techniques to enhance understanding and communication of dynamic pricing decisions under forecast uncertainty. It was found that lower tariffs are explained, for instance, by periods of higher renewable energy availability and lower market prices and that the most important feature was the hour of the day.

Abstract
This study presents the development and testing of satellite antennas for the SONDA probe, an innovative deepsea monitoring system designed to be deployed by high-altitude balloons. The probe descends to the deep ocean, resurfaces, and transmits data while functioning as a drifter. The project faced unique design constraints, including the need for low-cost materials and lightweight construction for balloon deployment. These constraints ruled out traditional hermetic housings, necessitating alternative solutions for antenna protection. The work focused on custom ceramic patch antennas and their performance under various protective coatings, which affected the antennas' resonance and gain. Thinner layers effectively protected the antennas from high-pressure conditions and water ingress, maintaining functionality. Experiments on antenna height revealed optimal positioning above the water surface to minimize wave-induced signal interference. Hyperbaric chamber tests validated the mechanical integrity and functionality of the antennas under pressures equivalent to depths of 1500 m Antenna characterization techniques were employed in an anechoic chamber to validate antenna performance with the coating and to assess their correct operation after the hyperbaric tests. Field deployments demonstrated the antennas' capability to transmit data after diving. Challenges included communication delays, corrupted data, and mechanical vulnerabilities in materials. The findings emphasize the importance of rigorous mechanical design, material selection, and system optimization to ensure reliability in marine environments. This work advances the development of low-cost, lightweight, and modular probes for autonomous ocean monitoring, with potential applications in long-term drifter studies, real-time marine monitoring and oceanographic research.

Abstract
Hyperparameter optimization (HPO) is critical for enhancing the predictive performance of machine learning models in credit risk assessment for peer-to-peer (P2P) lending. This study evaluates four HPO methods, Grid Search, Random Search, Hyperopt, and Optuna, across four models, Logistic Regression, Random Forest, XGBoost, and LightGBM, using three real-world datasets (Lending Club, Australia, Taiwan). We assess predictive accuracy (AUC, Sensitivity, Specificity, G-Mean), computational efficiency, robustness, and interpretability. LightGBM achieves the highest AUC (e.g., 70.77% on Lending Club, 93.25% on Australia, 77.85% on Taiwan), with XGBoost performing comparably. Bayesian methods (Hyperopt, Optuna) match or approach Grid Search’s accuracy while reducing runtime by up to 75.7-fold (e.g., 3.19 vs. 241.47 min for LightGBM on Lending Club). A sensitivity analysis confirms robust hyperparameter configurations, with AUC variations typically below 0.4% under ±10% perturbations. A feature importance analysis, using gain and SHAP metrics, identifies debt-to-income ratio and employment title as key default predictors, with stable rankings (Spearman correlation > 0.95, p<0.01) across tuning methods, enhancing model interpretability. Operational impact depends on data quality, scalable infrastructure, fairness audits for features like employment title, and stakeholder collaboration to ensure compliance with regulations like the EU AI Act and U.S. Equal Credit Opportunity Act. These findings advocate Bayesian HPO and ensemble models in P2P lending, offering scalable, transparent, and fair solutions for default prediction, with future research suggested to explore advanced resampling, cost-sensitive metrics, and feature interactions.

Abstract
This study explores the application of machine learning algorithms to optimize the geometry of the plasmonic layer in a surface plasmon resonance photonic crystal fiber sensor. By leveraging the simplicity of linear regression ( LR) alongside the advanced predictive capabilities of the gradient boosted regression (GBR) algorithm, the proposed approach enables accurate prediction and optimization of the plasmonic layer's configuration to achieve a desired spectral response. The integration of LR and GBR with computational simulations yielded impressive results, with an R-2 exceeding 0.97 across all analyzed variables. Moreover, the predictive accuracy demonstrated a remarkably low margin of error, epsilon < 10(-15). This combination of methods provides a robust and efficient pathway for optimizing sensor design, ensuring enhanced performance and reliability in practical applications.

Abstract
Crowdsourced data streams are popular and extremely valuable in several domains, namely in tourism. Tourism crowdsourcing platforms rely on past tourist and business inputs to provide tailored recommendations to current users in real time. The continuous, open, dynamic and non-curated nature of the crowd-originated data demands specific stream mining techniques to support online profiling, recommendation, change detection and adaptation, explanation and evaluation. The sought techniques must, not only, continuously improve and adapt profiles and models; but must also be transparent, overcome biases, prioritize preferences, master huge data volumes and all in real time. This article surveys the state-of-art of adaptive and explainable stream recommendation, extends the taxonomy of explainable recommendations from the offline to the stream-based scenario, and identifies future research opportunities.