Vera Miguéis

Abstract
In today's highly competitive fashion retail market, it is crucial to have accurate demand forecasting systems, namely for new products. Many experts have used machine learning techniques to forecast product sales. However, sales that do not happen due to lack of product availability are often ignored, resulting in censored demand and service levels that are lower than expected. Motivated by the relevance of this issue, we developed a two-stage approach to forecast the demand for new products in the fashion retail industry. In the first stage, we compared four methods of transforming historical sales into historical demand for products already commercialized. Three methods used sales-weighted averages to estimate demand on the days with stock-outs, while the fourth method employed an Expectation-Maximization (EM) algorithm to account for potential substitute products affected by stock-outs of preferred products. We then evaluated the performance of these methods and selected the most accurate one for calculating the primary demand for these historical products. In the second stage, we predicted the demand for the products of the following collection using Random Forest, Deep Neural Networks, and Support Vector Regression algorithms. In addition, we applied a model that consisted of weighting the demands previously calculated for the products of past collections that were most similar to the new products. We validated the proposed methodology using a European fashion retailer case study. The results revealed that the method using the Expectation-Maximization algorithm had the highest potential, followed by the Random Forest algorithm. We believe that this approach will lead to more assertive and better-aligned decisions in production management.

Abstract
Implementation of information and communication technologies (ICTs) in education is defined as the incorporation of ICTs into teaching and learning activities, both inside and outside the classroom. Despite widely studied, there is still no consensus on how it affects student performance. However, before evaluating this, it is crucial to identify which factors impact students' use of ICT for educational purposes. This understanding can help educational institutions to effectively implement ICT, potentially improving student results. Thus, adapting the conceptual framework proposed by Biagi and Loi (2013) and using the 2018 database of the Program for International Student Assessment (PISA) and a decision tree classification model developed based on CRISP-DM framework, we aim to determine which socio-demographic factors influence students' use of ICT for educational purposes. First, we categorized students according to their use of ICT for educational purposes in two situations: during lessons and outside lessons. Then, we developed a decision tree model to distinguish these categories and find patterns in each group. The model was able to accurately distinguish different levels of ICT adoption and demonstrate that ICT use for entertainment and ICT access at school and at home are among the most influential variables to predict ICT use for educational purposes. Moreover, the model showed that variables related to teaching best practices of Internet utilization at school are not significant predictors of such use. Some results were found to be country-specific, leading to the recommendation that each country adapts the measures to improve ICT use according to its context.

Abstract
Data-driven models have been recognized as powerful tools to support Structural Health Monitoring (SHM). This paper contributes to the literature by exploring two data-driven approaches to detect damage through changes in a set of variables that assess the condition of the structure, and accommodates the challenge that may arise due to the influence of environmental and operational variabilities. This influence is reflected in the response of the structure and can reduce the probability of detecting damage in a structure or increase the probability of signaling false positives. This paper conducts a comparative study between a machine learning detection approach (supported by linear regression, random forest, support vector machine, and neural networks) and a cointegration approach, with the aim of detecting damage as early as possible. This study also contributes to the literature by evaluating the merits of the damage detection methods using real data collected from a small-scale structure. The structure is analyzed in a reference state and a perturbed state in which damage is emulated. The results show that both approaches are able to detect damage within the first 24 h, without ever signaling false positives. The cointegration based approach can notably detect damage after 10 h and 15 minutes, while the machine learning approach takes 20 h 30 m to detect damage.

Abstract
Citizen engagement with Sustainable Energy Solutions (SES) is considered essential for the current energy transition, since decarbonization requires individuals to shift from passive consumers to citizens actively involved with the energy system. However, citizen engagement research has remained peripheral and scattered, particularly in what regards the drivers of engagement behaviors. To address this challenge, this study examines how different forms of perceived value of SES (utilitarian, social, and environmental) influence different types of citizen engagement behaviors (information seeking, proactive managing, sharing feedback, helping other users, and advocating). To this end, we developed a quantitative study in the context of a H2020 EU project, with a sample of 456 citizens from the city of Alkmaar (the Netherlands). Our findings show that the utilitarian value of SES has a significant effect on all the engagement behaviors, except for sharing feedback. Social value has a significant influence on the more socially related engagement behaviors, such as sharing feedback, helping other users, and advocating. Finally, environmental value has an indirect effect on information seeking, proactive managing, and advocating, but only when mediated through awareness of consequences. The implications of this study should allow SES providers to design more relevant offerings and policymakers to develop better citizen engagement strategies.

Abstract
Food waste is responsible for severe environmental, social, and economic issues and therefore it is imperative to prevent or at least minimize its generation. The main cause of food waste is poor demand forecasting and so it is essential to improve the accuracy of the tools tasked with these forecasts. The present work proposes four models meant to help food catering services predict food demand accurately and thus avoid overproducing or underproducing. Each model is based on a different machine learning technique. Two baseline models are also proposed to mimic how food catering services estimate future demand and to infer the added value of employing machine learning in this context. To verify the impact of the proposed models, they were tested on data from the three different canteens chosen as case studies. The results show that the models based on the random forest algorithm and the long short-term memory neural network produced the best forecasts, which would lead to a 14% to 52% reduction in the number of wasted meals. Furthermore, by basing their decisions on these forecasts, the food catering services would be able to reduce unmet demand by 3% to 16% when compared with the forecasts of the baseline models. Thus, employing machine learning to forecast future demand can be very beneficial to food catering services. These forecasts can increase the service level of food services and reduce food waste, mitigating its environmental, social, and economic consequences.