NIMS - Teses de Doutoramento (Doctoral Theses)
URI permanente para esta coleção:
Navegar
Entradas recentes
- Enhancing Ecosystem-based Disaster Risk Reduction Through Geoinformatics: Integrating Ecological Factors into Assessment PracticesPublication . Broquet, Melanie; Cabral, Pedro da Costa Brito; Campos, Felipe Siqueira eEcosystem-based Disaster Risk Reduction (Eco-DRR) leverages ecosystem conservation and restoration to mitigate natural disaster risks by providing regulatory services that reduce their intensity and lower the vulnerability of exposed communities and ecosystems. Despite growing recognition, especially after the 2004 Indian Ocean tsunami, Eco-DRR faces key barriers: limited consensus on ecosystem effectiveness, inconsistent methodologies, data scarcity, and the absence of standardized frameworks that integrate ecological factors into disaster risk assessments. This study addresses these gaps by using landslide hazard as a case study to strengthen the credibility and acceptance of Eco-DRR through empirical evidence and improved assessment practices. Three inter-connected objectives guided the work: (1) identifying ecological factors relevant for landslide susceptibility assessment (LSA) through literature review; (2) examining the relationship between Land Use/Land Cover (LULC) and habitat quality as indicators of hazard-prone landscapes using the InVEST Habitat Quality model; and (3) testing whether integrating ecological variables into LSA frameworks improves predictive performance using Random Forests. Findings show that eco-environmental factors, especially dynamic ones, remain underutilized in LSAs. LULC change was strongly correlated with ecological degradation, supporting the use of integrative indicators such as habitat quality for characterizing vulnerable landscapes. Integrated models combining structural and ecological variables, particularly dynamic ones, significantly outperformed conventional LSA models. These results confirm that eco-environmental variables play a critical role in shaping landslide susceptibility and should be systematically integrated into risk assessments. Overall, this research strengthens Eco-DRR’s scientific foundation by moving beyond static, hazard-centric approaches. It introduces evidence-based methodologies that are applicable even in data-scarce contexts, replicable across settings. By promoting cross-disciplinary integration and efficiency, the work helps bridge the knowledge–action gap, enhances policy relevance, and underscores ecosystems as critical assets whose protection and restoration are essential for breaking the cycle of degradation and disaster risk.
- Understanding the drivers of academic achievement: A multi-method approachPublication . Afonso, Ana Beatriz Antunes; Jesus, Frederico Miguel Campos Cruz Ribeiro deEducation is a fundamental driver of social mobility, sustainable development, and economic growth. Yet, despite its importance, the determinants of academic achievement (AA) remain contested, with fragmented findings often based on limited sample data. This work addresses this gap by leveraging administrative records from virtually all Portuguese public high school students, complemented by a targeted survey, to provide a comprehensive and context-sensitive analysis of AA. Rather than relying on isolated case studies, this thesis follows a multi-phase design that progressively deepens our understanding of AA. It begins by mapping global evidence on AA drivers, then examines virtually every student in the Portuguese high school system to assess how unprecedented eventssuch as the COVID-19 pandemic and regional disparities shape student outcomes. Finally, it integrates primary data to uncover how family environments, in particular parental involvement, interact with socioeconomic conditions during the transition to higher education. Several consistent findings emerged across the studies. Among those, socioeconomic status proved to be one of the strongest predictors of AA. The COVID-19 pandemic deepened existing inequities and altered the relative importance of different success factors. Regional disparities also became evident, with rural and urban students demonstrating distinct needs linked to unequal access to resources. Finally, parental involvement played a crucial role, not only exerting a direct influence on student outcomes but also moderating the effects of socioeconomic conditions. This thesis delivers a unique, comprehensive, large-scale analysis of AA in Portugal using the entire public secondary school population. It demonstrates the added value of machine learning over traditional methods in handling large-scale educational data. It generates actionable insights aligned with international policy frameworks such as the United Nations’ fourth Sustainable Development Goal. These contributions advance theoretical understanding while offering practical guidance for the design of more inclusive and equitable education systems.
- Advanced Machine Learning for Building Energy Efficiency: Integrating Predictive Models and Optimization Techniques for Smarter Energy Management and Strategic DecisionsPublication . Almeida, Fernando Pedro Silva; Castelli, Mauro; Côrte-Real , Nadine Evangelista de PinhoBuilding operations account for a significant portion of global energy consumption and carbon emissions, with heating, ventilation, and air conditioning (HVAC) systems being among the most energy-intensive components. Addressing inefficiencies in HVAC operations is crucial for achieving energy savings, reducing greenhouse gas emissions, and meeting sustainability goals, such as those outlined in the Paris Agreement, the EU’s Energy Performance of Buildings Directive (EPBD), and the REPowerEU plan. This study presents a comprehensive framework for forecasting, optimizing, and controlling space heating and cooling energy consumption in buildings using a combination of machine learning (ML), deep learning (DL), and reinforcement learning (RL) techniques. The research evaluates a broad range of ML and DL algorithms, including XGBoost, Random Forest, Support Vector Regression, Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU), and Transformer models, for their effectiveness in predicting space heating and cooling loads. These models are trained on real-world operational and meteorological data collected from the European Central Bank (ECB) building and demonstrate improved accuracy over traditional methods, mainly when feature selection includes localized weather conditions and building-specific variables. For space heating consumption, XGBoost achieved an R² of 0.966. At the same time, Random Forest consistently outperformed other models in cooling load prediction across multiple system types, such as Cooling Ceiling, Cooling Ventilation, Free Cooling, and Total Cooling. In addition to predictive modeling, the study develops a recommendation system based on LSTM networks to support energy efficiency decisions across space heating and selected cooling systems. The system provides hourly and daily forecasts, allowing energy managers to adjust operations dynamically. To enhance interpretability, OpenAI’s GPT4 model is integrated to offer contextual explanations of time-series graph outputs, facilitating non-technical understanding and informed decision-making. The feasibility, effectiveness, accuracy, and reliability of this natural language–based operational decision support and recommendation component are intended to be evaluated through enduser studies, which are identified as part of future work. The work further addresses the balance between energy efficiency and thermal comfort by utilizing DL-based forecasting models for multi-zone temperature and space heating consumption. A two-stage optimization process ensures occupant comfort (typically within 21–23°C in winter) while minimizing heating energy use, with LSTM and Transformer models achieving up to 21% reduction in heating demand compared to actual consumption, as calculated by comparing optimized heating schedules against historical baseline energy use over the same period. This reduction is validated across multiple building zones and weather conditions, confirming the effectiveness of the predictive optimization framework. Finally, a deep reinforcement learning framework is proposed to enable real-time adaptive control of HVAC systems. Among the tested models, Deep Deterministic Policy Gradient (DDPG) achieved the lowest comfort violations and temperature instability, outperforming TD3 and D-SAC, and demonstrating strong potential for practical deployment in Building Energy Management Systems (BEMS). This study contributes robust, scalable, and interpretable methodologies for data-driven energy management in buildings. It provides actionable solutions that not only enhance forecasting and operational efficiency but also support broader environmental and institutional objectives in the context of innovative, sustainable building design and management.
- Project Governance: Examining the Role of PMOs and PMIS Impact in Enhancing Organizational Project ManagementPublication . Monteiro, António José Vieira Póvoa; Santos, Vítor Manuel Pereira Duarte dos; Varajão, João Eduardo Quintela Alves de SousaOrganizational Project Management (OPM) has become essential for effective project governance by aligning strategic goals with project execution, particularly through organizational structures such as Project Management Offices (PMOs). However, the diversity of PMO configurations and their evolving role within increasingly complex and digitalized environments present significant challenges for both theory and practice. This research explores the contribution of PMOs to OPM by analyzing their typologies, functions, and their interaction with project management information systems (PMIS). The investigation unfolds in three phases. The first phase involved a systematic review of the literature to consolidate existing PMO typologies and types, identifying 16 typologies and 60 distinct PMO types. This phase revealed the dynamic and heterogeneous nature of PMOs and the lack of consensus regarding their functional boundaries. The second phase examined the evolution of PMO functions over the past two decades. Through literature analysis and a case study in a large organization, this phase confirmed the continued relevance of core PMO functions, such as monitoring performance and standardizing methodologies, while identifying emerging functions including stakeholder management, vendor coordination, and operational excellence in AI-driven projects. In the third phase, the research addressed the intersection between organizational structures and project tools. A conceptual model was developed and tested to evaluate the moderating role of PMOs in the relationship between PMIS use and project manager performance. Structural equation modeling (SEM) was conducted to assess the validity and reliability of the measurement model and to evaluate the hypothesized structural relationships within the conceptual model. The findings indicate the importance of establishing well-structured project management practices with technology, ensuring seamless alignment between organizational processes and technology. The findings also indicate that the presence of PMOs strengthen the effectiveness of PMIS, enhancing project decision-making and execution. Overall, this research contributes to the understanding of OPM by offering a consolidated and updated view of PMO configurations and functions, while also highlighting the importance of aligning governance structures and digital systems to improve organizational project outcomes.
- Enhancing Geometric Semantic Genetic Programming: A Study on Dynamic Population and Deflate Geometric MutationPublication . Farinati, Davide; Vanneschi, LeonardoGenetic Programming (GP) is a branch of evolutionary computation that evolves programs to solve complex, non-linear problems without requiring predefined models. A key application of GP is symbolic regression, which autonomously discovers mathematical expressions that best represent data patterns. However, standard GP operators create a rugged fitness landscape, making optimization challenging. To address these limitations, Geometric Semantic Genetic Programming (GSGP) was introduced, employing Geometric Semantic Operators (GSOs) that induce a unimodal fitness landscape, enhancing performance. However, GSGP suffers from excessive bloat, transforming it into a black-box model. This dissertation presents several advancements to mitigate this issue. First, a dynamic population approach is adapted to GSGP, optimizing fitness evaluations and reducing computational cost. Next, a novel algorithm, Semantic Learning algorithm based on Inflate and deflate Mutation (SLIM), is introduced, utilizing a new Deflate Geometric Semantic Mutation (DGSM) operator that generates smaller offspring while preserving an unimodal fitness landscape. Experimental results show that SLIM produces significantly smaller models than GSGP, with performance comparable to or better than traditional GP. A Python library implementing SLIM is also implemented. Further studies explore SLIM’s characteristics, including the impact of DGSM on overfitting and feature selection. A comprehensive analysis of the Mutation Step (MS) parameter is also conducted. Additionally, new Geometric Semantic Crossovers (GSCs) are employed to enhance SLIM’s evolutionary capabilities. Finally, SLIM is applied to genomic data for predicting drug response, showing SLIM advantage in real use case scenarios. This research advances genetic programming by improving interpretability, reducing computational costs, and expanding applications in data-driven modeling.
- Contextual Analysis of Case Law: Enhancing Precedent Identification with Artificial IntelligencePublication . Silva, Hugo Saisse Mentzingen da; António, Nuno Miguel da Conceição; Bação, Fernando José Ferreira LucasIn the field of Legal Informatics and Artificial Intelligence, this research addresses the challenge of efficiently identifying relevant legal precedents and understanding their context, an issue stemming from the current lack of effective end-to-end automated retrieval methods combined with visual analytics and contemporary AI assistance. To fill this gap, the study explores, evaluates, and integrates state-of-the-art language models and text embeddings to enhance precedent identification and contextual analysis of case law. The research encompassed a systematic literature review to establish the foundations and survey empirical methods for automated precedent retrieval. Further experiments employed text summarization and large language models to increase precedent identification performance and reduce processing time and costs. In parallel, a review of visual analytics of legal corpora informed the design of interactive visualizations to contextualize decisions from an administrative court. All these insights were integrated into a prototype application called JurisMap, which combines advanced NLP with intuitive visual interfaces, a chatbot and AI agents to assist legal professionals in precedent research and administrative defense preparation. The results demonstrate that classical and neural models each offer complementary strengths across different performance metrics, that summarization significantly improves retrieval efficiency, and that visualization enhances the interpretability of case law context. Ultimately, this research contributes a comprehensive, experimentally validated, and practically deployable framework for AI-assisted precedent retrieval and contextual analysis, effectively bridging theoretical developments in Legal Informatics with real-world legal practice.
- Understanding and facilitating SNS use reduction: A multi-method investigation of psychological and behavioral mechanismsPublication . Nascimento, Pedro Eduardo Facho do; Oliveira, Tiago André Gonçalves Félix de; Neves, Joana Paisana Pires Costa dasIn the digital age, social networking sites (SNS) have become deeply embedded in the fabric of everyday life, offering substantial social, informational, and emotional benefits. However, growing concerns about excessive and compulsive use have brought to light the darker side of these platforms, particularly their impact on mental health and overall well-being. Amid rising calls for more mindful technology use, SNS use reduction has emerged as a promising yet underexamined strategy for mitigating the harms associated with hedonic information systems (IS). This work seeks to advance a nuanced understanding of the SNS use reduction process by drawing on multiple theoretical lenses and methodological approaches. It begins with a systematic literature review, synthesizing the fragmented body of knowledge surrounding hedonic IS use reduction and identifying key gaps in theory and practice. Building on this foundation, the research adopts a cognitive-affective framework to examine how individuals' emotional responses and rational evaluations jointly influence the motivation to reduce SNS engagement. Incorporating a risk- and fear-based perspective, the thesis further unpacks the complex interplay between contextual triggers (e.g., perceived vulnerability), organic responses (e.g., fear of adverse consequences), behavior response and the consequence of reduced use. A mixed-methods study highlights the role of distractions- both internal and external to SNS - as mechanisms that inadvertently impact the disengagement process. Additionally, this work explores how a neurodiversity condition - attention deficit hyperactivity disorder (ADHD) - poses unique challenges to use reduction, particularly through impaired self-regulation and time management capacities. Collectively, the five studies presented in this thesis illuminate the multifaceted nature of SNS use reduction, revealing it as a dynamic process shaped by psychological, contextual, and cognitive-behavioral dimensions. The findings not only deepen theoretical insight into IS self-regulation but also inform practical interventions aimed at promoting healthier relationships with technology. Ultimately, this work contributes to ongoing efforts to foster more intentional and balanced digital engagement in a hyperconnected world.
- Spatiotemporal Characterization of Land Cover with Remote Sensing: Exploring Training Data Strategies and Sentinel-2 Time SeriesPublication . Moraes, Daniel; Caetano, Mário Sílvio Rochinha de Andrade; Campagnolo, Manuel Lameiras de FigueiredoThis thesis explores two key areas at the intersection of land cover and remote sensing: land cover mapping and continuous monitoring. While using machine learning for land cover mapping has become widespread, a notable gap in the literature exists regarding the organization and synthesis of training data approaches. Furthermore, there is a lack of studies exploring methods to reduce the heavy dependence of state-of-the-art techniques, such as deep learning, on training data. On the continuous monitoring side, although several studies have used Landsat or harmonized Landsat-Sentinel-2 data, there is a limited focus on utilizing Sentinel-2 data alone for continuous land cover monitoring. To address these gaps, this thesis proposes a series of studies to answer specific research questions related to land cover mapping and monitoring. A systematic review is conducted to synthesize how training data has been addressed in the remote sensing literature. Existing point-based reference land cover data is integrated with weakly and self-supervised learning for national scale land cover mapping in Portugal. Lastly, the feasibility of using Sentinel-2 data for continuous forest loss monitoring with the Continuous Change Detection and Classification algorithm is assessed in a dynamic landscape in Portugal. The research results include a comprehensive synthesis of the various methods used for training data in land cover mapping, providing a valuable guide for future research. Additionally, the thesis demonstrates the effectiveness of combining deep learning with weakly and self-supervised learning and leveraging reference datasets to reduce the need for fully annotated training data. Finally, the research shows that Sentinel-2 time series can enable accurate, spatially detailed, and agile monitoring of forest loss, showcasing its potential for continuous land cover monitoring. These findings contribute to advance the knowledge in the field by opening new avenues for more efficient and scalable land cover mapping and monitoring.
- Developing The Firm Under the Innovation Paradigm: A vision of Industry 4.0 in European industries, Industry 4.0 maturity measurement, and integration-related innovativenessPublication . Branco, Maria Isabel Cabral de Abreu Castelo; Oliveira, Tiago André Gonçalves Félix de; Jesus, Frederico Miguel Campos Cruz Ribeiro deThe pace of technological development in the last quarter of the XX century and the first quarter of the XXI century has made technological innovation a source of competitive advantage. While some firms have been able to easily recognize this fact, several others have been slower in grasping all the managerial challenges that come with a world more interconnected, more virtualized, and where data may become valuable information. Industry 4.0 is a term that originated in the context of German Industrial policy in the second decade of the XXI century, looking to encapsulate the possibilities offered to manufacturing by certain emerging technologies like IoT, CPS, CC, AM, or BDA. As a label that is part of an institutional discourse that associates technological innovation and the outcomes of the 4th Industrial Revolution, quite rapidly became the common denominator of industrial policies in several European countries. The application of the new technologies to production processes results in interconnectivity, interoperability, virtualization, collaboration, and information transparency at the factory level but also impacts the relationships with suppliers and customers. Available data from the Eurostat showed three different dimensions that characterize Industry 4.0 implementation across the European Union: Industry 4.0 Infrastructure, Big Data Maturity, and Industry 4.0 Applications. It also showed five profiles, according to the relative distribution of each dimension: Infrastructure Focused, Average -> Laggards, Infrastructure Leaders, Big Data Leaders, and Applications Leaders. The combination of the dimensions and the profiles allowed for the characterization of Industry 4.0 implementation across countries and economic sectors. The sector, more than the country, better explains the Industry 4.0 divide across Europe. At the individual firm level, however, a more detailed assessment of the degree of Industry 4.0 implementation demands an integrated overview, beyond the pace of adoption of new technologies. Because integrating these new technologies into productive processes generates new types of operations, new value propositions, and new ways of creating and managing relationships with suppliers and customers, the whole value chain is impacted. As such, Industry 4.0 also requires the firm to recognize the need to develop some capabilities that allow it to adapt and transform and to be able to manage knowledge and technological resources. Therefore, to understand the degree of maturity of Industry 4.0 implementation at the firm level, and to allow for cross-firm comparisons on the maturity level of that implementation, an integrated perspective must consider the value chain manifestations and the availability of the necessary enablers. Also at the individual firm level, higher integration with suppliers and customers is made possible by new technologies or more collaborative business practices. Under the ecosystem paradigm, firms commit their specific resources to develop complex, multi-contributed value propositions, together with other firms, forming collaborative relationships that go beyond the linear supply chain and that are studied under the ecosystem framework. Absorptive capacity is a resource firms may apply to deepen integration with suppliers and customers. Innovativeness, or the propensity of the firm to innovate, is positively impacted by customer integration and proactivity in understanding customers’ needs. Furthermore, innovativeness positively influences the ability of firms to implement changes in the value-capturing mechanisms that constitute their business model, which ultimately allows capturing value from the ecosystem’s activities.
- Firm-level effects of the Portuguese R&D tax credit: a microeconomic and sectoral analysisPublication . Paredes, José Alexandre da Silva; Damásio, Bruno Miguel PintoThis Doctoral thesis investigates the impact of innovation policies, focusing on the Oslo Manual's role in shaping innovation research and the effects of R&D tax credits on firm behaviour and employment dynamics. Addressing gaps in the literature, it explores the Oslo Manual's academic influence, the allocation of highly qualified personnel in response to R&D tax credits, and the broader employment effects of these incentives. The first study (Chapter 3), "Accounting for the Oslo Manual: reflecting on the past and setting the stage for future research", applies bibliometric and textmetric analyses to over 1,300 research articles, assessing the Oslo Manual's adoption and relevance over three decades. The findings highlight its increasing importance, particularly after 2008, and its integration with fields such as entrepreneurship, performance, and knowledge management. The second study (Chapter 4), "Does R&D tax credit impact firm behaviour? Micro evidence for Portugal", investigates how R&D tax credits influence the allocation of PhD holders across firms with different R&D intensities. Using firm-level data (1995–2017) from Portugal, the study finds that tax credits significantly affect the distribution of PhD holders in medium-high and high R&D intensity firms after three years. This research shifts the focus from R&D expenditure to human capital effects. The third study (Chapter 5), "Do R&D tax credits really boost hiring? Insights into employment dynamics", employs a difference-in-differences approach with a staggered design to assess employment effects in Portugal (2014–2022). Results indicate a substantial increase in R&D staff, with sector-specific variations, such as an 18.4% rise in information and communication and 12.3% in manufacturing. This thesis advances knowledge by bridging theoretical and empirical perspectives on science and technology policies. It provides insights into the Oslo Manual's influence, the human capital effects of R&D tax credits, and sectoral employment dynamics, offering evidence-based recommendations for policymakers and future research.