https://hdl.handle.net/10216/6812 2026-08-07T14:12:27Z https://hdl.handle.net/10216/176181 Title: Data Centric Wound Tissue Segmentation Abstract: Chronic wounds represent a global health concern affecting millions, often resulting from diabetes, venous disorders, or immobility. These wounds place a significant burden on healthcare systems and reduce patient quality of life due to complex diagnostic and treatment challenges. Traditional assessment relies heavily on subjective visual evaluation, leading to frequent inconsistencies. Since precise wound segmentation is essential for objective clinical assessment and treatment monitoring, this study presents an automated pipeline for wound and tissue-level segmentation. Rather than comparing novel architectures, this research answers the question how different data splitting strategies affect training stability, convergence speed, task generalization, and overall performance on unseen data. Five distinct, wound-aware splitting methods were employed to prevent data leakage, ranging from a traditional random split and multidimensional stratification (based on phototype, typology, and tissue presence) to advanced batch-aware clustering and deep visual feature-based (FSCIS) splitting. Each strategy was evaluated across three architectures: a lightweight CNN (DeepLabV3-ResNet50) and two transformer-based models (UPerNet-Swin and Mask2Former-Swin-L). Following 3-fold cross-validation hyperparameter tuning, final models were trained on an 80/20 split and evaluated on a fixed holdout test set. Experimental results demonstrate that advanced data-centric splits, particularly Batch-Aware and FSCIS, improve training stability and convergence while delivering superior performance. In binary wound segmentation, UPerNet-Swin paired with FSCIS achieved the highest result (90.64% Wound-DSC), followed closely by the Batch-Aware split (90.59% Wound-DSC). For the more challenging tissue-level segmentation, the Batch-Aware strategy combined with UPerNet-Swin reached 69.87% Avg3-DSC, again offering the best balance of performance and stability. In contrast, the baseline random split resulted in the lowest performance (for DeepLabV3-R50 at 63.76% of Avg3-DSC), while the over-stratified Multidim3 split caused serious generalization gaps, with performance falling to 61.80% Avg3-DSC for Mask2Former-Swin-L. These findings confirm that matching data distribution to the model architecture is essential for achieving reliable and objective wound assessment in clinical practice. This study provides a robust framework for data splitting, demonstrating that even marginal gains in model reliability have a significant impact on clinical decision-making. Description: This dissertation investigates and compares the effects of various dataset splitting strategies on segmentation models for chronic wound and tissue detection. 2026-07-23T00:00:00Z https://hdl.handle.net/10216/176179 Title: A Framework for Evaluating the Correctness of Programming Learning Platforms Feedback Abstract: The test oracle problem has gained renewed attention with the emergence of Large Language Models (LLMs), which introduce both opportunities and challenges for automated software testing. We need to understand to what extent problem specifications can be used as oracles to guide test case generation, and whether it is possible to overcome initial difficulties. This analysis is conducted on an online academic platform. A five-step workflow is proposed: (1) collecting problem solutions and their specifications and verifying their acceptance by the platform; (2) generating test cases from the specifications; (3) deriving pre- and post-conditions from the specifications to produce a formal specification; (4)evaluating the performance of the initial prompt across other problem categories; and (5) refining the prompt to achieve improved results. Together, these steps enable a thorough assessment of the effectiveness of this approach. A review of 207 Python solutions revealed that a small fraction, roughly 1.93% (4 out of 207), were mistakenly accepted by the platform despite being incorrect. Of all test cases generated, 1,778 were deemed valid, accounting for 90.99% of the total output. Among the 207 challenges evaluated, 58 (28.0%) contained at least one failing test case, with some challenges accumulating up to 12 failures. Notably, 13 challenges achieved a 0% success rate, meaning every single generated test case failed. Finally, through prompt refinement, the Test Execution Success Rate improved substantially, rising from 54% to 70.9%. In conclusion, this dissertation explores the effectiveness of leveraging problem specifications as test oracles, and assesses whether observed shortcomings, such as high failure rates in certain challenges, can be mitigated by incorporating them as additional context to enhance oracle generation. Description: A five-step workflow is proposed: (1) collecting problem solutions and their specifications and verifying their acceptance by the platform; (2) generating test cases from the specifications; (3) deriving pre- and post-conditions from the specifications to produce a formal specification; (4)evaluating the performance of the initial prompt across other problem categories; and (5) refining the prompt to achieve improved results. Together, these steps enable a thorough assessment of the effectiveness of this approach. 2026-07-21T00:00:00Z https://hdl.handle.net/10216/176178 Title: Simulation and Reinforcement Learning for Decision Support in Remanufacturing Networks 2026-07-17T00:00:00Z https://hdl.handle.net/10216/176177 Title: Capturing and Preserving Tacit Knowledge in Software Engineering Teams 2026-07-09T00:00:00Z https://hdl.handle.net/10216/176176 Title: An AI-Driven Autonomous Self-Healing System for Cloud-Native Microservice Applications 2026-07-21T00:00:00Z https://hdl.handle.net/10216/176175 Title: AI-Driven Wildfire Detection from UAV Imagery 2026-07-15T00:00:00Z https://hdl.handle.net/10216/176174 Title: UP2Speed - Perception system for monitoring and control of several autonomous racing vehicles and environment 2026-07-24T00:00:00Z https://hdl.handle.net/10216/176173 Title: Measuring What Matters: Individual Vs. Team Productivity in Software Development 2026-07-10T00:00:00Z https://hdl.handle.net/10216/176172 Title: Scaling Intelligence: A Framework for Designing Internal Processes in Technology-Intensive Startups 2026-07-17T00:00:00Z https://hdl.handle.net/10216/176171 Title: Product Validation Accelerator Engine - Combining AI with Software Testing Techniques for Configurable Insurance Systems Abstract: The validation of insurance product configurations is a complex and time-consuming process. Highly configurable insurance systems generate combinatorial spaces that grow exponentially with the number of available options, and the interdependencies between business rules, calculation formulas, and variation conditions make systematic coverage intractable through manual means. This dissertation proposes an approach to accelerate the validation of these configurations by combining Artificial Intelligence with traditional software testing techniques. In the proposed architecture, AI does not act as a generic orchestrator, but rather as the component capable of understanding the subtleties inherent to the domain: interpreting premium calculation formulas, analyzing composition rules and their variations, and identifying boundary conditions and critical scenarios that a purely deterministic pipeline cannot infer. This semantic knowledge is then combined with deterministic methods, including combinatorial generation, Boundary Value Analysis, and Equivalence Class Partitioning, to produce test scenarios that are valid, efficient, and relevant to the business. The proposed solution, named Product Validation Accelerator Engine, has significantly improved the validation time of product configurations by automating, or semi-automating, the three most demanding processes in the workflow: the creation of test artifacts, targeted verifications through unit tests, and mass validation through massive tests. The entire solution is adaptable to the configuration inherent to insurance systems and to the complexity they present. Description: A validação de configurações de produtos de seguros é um processo complexo e moroso. Os sistemas de seguros altamente configuráveis geram espaços combinatórios que crescem exponencialmente com o número de opções disponíveis, e as interdependências entre regras de negócio, fórmulas de cálculo e condições de variação tornam a cobertura sistemática por meios manuais muito desafiante. Esta dissertação propõe uma abordagem para acelerar a validação destas configurações, ao combinar Inteligência Artificial com técnicas tradicionais de teste de software. Na arquitetura proposta, a IA não atua como um orquestrador genérico, mas sim como a componente capaz de compreender as subtilezas inerentes ao domínio: interpretar fórmulas de cálculo de prémios, analisar regras de composição e respetivas variações, e identificar condições de fronteira e cenários críticos que um algoritmo puramente determinístico não consegue inferir. Este conhecimento semântico é então combinado com métodos determinísticos, incluindo a geração combinatória, Análise de Valores Fronteira e Partição de Classes de Equivalência, para produzir cenários de teste válidos, eficientes e relevantes para o negócio. A solução proposta, denominada Product Validation Accelerator Engine, melhorou significativamente o tempo de validação de configurações de produtos ao automatizar, ou semi-automatizar, os três processos mais exigentes do fluxo de trabalho: a criação de artefactos de teste, as verificações direcionadas através de testes unitários e a validação em massa através de testes massivos. Toda a solução é adaptável à configuração inerente aos sistemas de seguros e à complexidade que estes apresentam. 2026-07-20T00:00:00Z https://hdl.handle.net/10216/176170 Title: Enhancing Productivity in Viticulture: An Integrated Approach to Planning and Operational Management 2026-07-08T00:00:00Z https://hdl.handle.net/10216/176168 Title: Visibilidade e Controlo Operacional numa Transição de Estabelecimento Logístico: Um caso de estudo 2026-07-22T00:00:00Z https://hdl.handle.net/10216/176167 Title: Fostering Organic Composting: A Service Design Approach 2026-07-20T00:00:00Z https://hdl.handle.net/10216/176166 Title: Integrating Human Oversight in Automated Data Standardisation: the NOUS Project use case 2026-07-22T00:00:00Z https://hdl.handle.net/10216/176165 Title: UPWIND- Development of a Take-off and Landing Platform for Integration in an Airborne Wind Energy System 2026-07-24T00:00:00Z https://hdl.handle.net/10216/176164 Title: Phenotyping Obstructive Sleep Apnea: integration of clinical and biomedical data through clustering analysis 2026-07-15T00:00:00Z https://hdl.handle.net/10216/176163 Title: Robotic System for Arbutus Berry Harvesting 2026-07-24T00:00:00Z https://hdl.handle.net/10216/176162 Title: Controller for Brushless Motors using ARM 2026-07-24T00:00:00Z https://hdl.handle.net/10216/176161 Title: Redesigning a Factory Layout 2026-07-22T00:00:00Z https://hdl.handle.net/10216/176159 Title: Design and Evaluation of an Explainable Clinical Decision Support System for Depression-Related Triage Based on Digital Phenotyping 2026-07-17T00:00:00Z