Publicação
Fraud Detection Systems Empowered by Context-Awareness: Leveraging Dynamic Machine Learning Techniques
| datacite.subject.fos | Ciências Naturais::Ciências da Computação e da Informação | pt_PT |
| dc.contributor.advisor | Castelli, Mauro | |
| dc.contributor.advisor | Agostinho, Nuno Filipe Rosa | |
| dc.contributor.author | Alexandre, Daila Silva | |
| dc.date.accessioned | 2024-10-28T10:56:12Z | |
| dc.date.available | 2024-10-28T10:56:12Z | |
| dc.date.issued | 2024-10-25 | |
| dc.description | Dissertation presented as the partial requirement for obtaining a Master's degree in Data Science and Advanced Analytics, specialization in Data Science | pt_PT |
| dc.description.abstract | In the dynamic and rapidly evolving field of financial technology, effective fraud detection systems (FDS) are crucial for maintaining security and customer trust. Traditional FDS methods often rely on static models, which struggle to adapt to emerging fraud patterns and evolving data distributions. This thesis explores the application of dynamic techniques to optimize fraud detection results, moving beyond conventional risk scores and static thresholds. The research employs a combination of machine learning (ML) approaches, including the use of secondary models for dynamic threshold adjustment and reinforcement learning (RL) for direct decision-making. A robust methodology was adopted, starting with an in-depth understanding of the business context, data exploration, feature engineering, and advanced sampling strategies to address data imbalance. Various strategies were tested, including anomaly detection using Isolation Forests, dynamic threshold setting through meta-learning, and RL to eliminate traditional score-based decision processes. The results demonstrated improvements in maximizing the true positive rate (TPR) while maintaining the false positive rate (FPR), ensuring more accurate and reliable fraud detection. By leveraging real-world data from a European transaction processor, this study provides empirical evidence of the benefits of dynamic optimization techniques in practical applications. The findings contribute to the ongoing development of adaptive and efficient FDS, highlighting the potential of combining advanced ML and RL methods to enhance the robustness and effectiveness of FDSs. | pt_PT |
| dc.identifier.tid | 203782518 | pt_PT |
| dc.identifier.uri | http://hdl.handle.net/10362/174155 | |
| dc.language.iso | eng | pt_PT |
| dc.rights.uri | http://creativecommons.org/licenses/by/4.0/ | pt_PT |
| dc.subject | Fraud Detection | pt_PT |
| dc.subject | Fraud Detection Systems | pt_PT |
| dc.subject | Dynamic Systems | pt_PT |
| dc.subject | Machine Learning | pt_PT |
| dc.subject | Reinforcement Learning | pt_PT |
| dc.subject | SDG 9 - Industry, innovation and infrastructure | pt_PT |
| dc.subject | SDG 11 - Sustainable cities and communities | pt_PT |
| dc.title | Fraud Detection Systems Empowered by Context-Awareness: Leveraging Dynamic Machine Learning Techniques | pt_PT |
| dc.type | master thesis | |
| dspace.entity.type | Publication | |
| rcaap.rights | openAccess | pt_PT |
| rcaap.type | masterThesis | pt_PT |
| thesis.degree.name | Mestrado em Ciência de Dados e Métodos Analíticos Avançados, especialização em Ciência de Dados | pt_PT |