Browsing by Author "RUA, ANA CATARINA FARIA"
Now showing 1 - 1 of 1
Results Per Page
Sort Options
- Enhancing supervised learning robustness investigating the impact of label noise on algorithm performancePublication . RUA, ANA CATARINA FARIA; Rodrigues, Maria de Fátima CoutinhoSupervised learning serves as the foundation for many AI systems because it enables models to learn from labelled examples. However, label noise resulting from human annotation errors or systematic biases can diminish model performance and limit generalization capabilities. This challenge is particularly significant in critical domains such as healthcare, finance, and autonomous systems. This thesis focuses on studying the impact of label noise on supervised learning algorithms in order to evaluate its influence across different datasets and to propose robust strategies for mitigation. This project includes methods of loss correction, data augmentation, and advanced noise detection frameworks as examples and demonstrates their prospective advantages through empirical experiments. The provided noise-robust algorithms in the research can be used with any real-world scenarios to improve the resilience of the algorithm. The findings are meant to be a connection between academic research and practical implementation by offering guidelines for handling noisy datasets effectively while ensuring model reliability and fairness. The proposed approach raised the average F1-Score from 0.647 under baseline conditions to 0.757 after full optimization.