Fraud Detection with Positive and Unlabeled Dataset using Interactive Learning Technique

Mobin Al HASSAN

Master's thesis

Abstract:

This thesis focuses on fraud detection using Positive and Unlabeled (PU) datasets with interactive learning approach, addressing the challenge of training machine learning models with messy, unlabeled real-world data, a common example is financial transactions where labeling is expensive or impossible. Unlike traditional well-balanced datasets, PU learning deals with a few positive examples and a large set of unlabeled data. The research proposes using a treebased classifier for tabular datasets in conjunction with a reinforcement algorithm instead of the commonly used neural networks in existing algorithms. This study aims to improve the accuracy and efficiency of fraud detection systems by leveraging interactive learning techniques, thereby overcoming significant data quality and labeling challenges in real-world scenarios. The results will provide insights into the practical application of these techniques, highlighting their effectiveness in enhancing fraud detection systems.

Abstract:

This thesis focuses on fraud detection using Positive and Unlabeled (PU) datasets with interactive learning approach, addressing the challenge of training machine learning models with messy, unlabeled real-world data, a common example is financial transactions where labeling is expensive or impossible. Unlike traditional well-balanced datasets, PU learning deals with a few positive examples and a large set of unlabeled data. The research proposes using a treebased classifier for tabular datasets in conjunction with a reinforcement algorithm instead of the commonly used neural networks in existing algorithms. This study aims to improve the accuracy and efficiency of fraud detection systems by leveraging interactive learning techniques, thereby overcoming significant data quality and labeling challenges in real-world scenarios. The results will provide insights into the practical application of these techniques, highlighting their effectiveness in enhancing fraud detection systems.

Keywords

PU Learning, weakly supervised learning, semi-supervised learning, classification, Imbalanced data, Data Science, Artificial Intelligence

Jazyk práce: English

Datum vytvoření / odevzdání či podání práce: 20. 8. 2024

Obhajoba závěrečné práce

• Vedúci: prof. Dr. Helena Liebelt

UNIVERSITY OF SOUTH BOHEMIA IN ČESKÉ BUDĚJOVICE

Faculty of Science

Master programme / odbor:
Artificial Intelligence and Data Science / Artificial Intelligence and Data Science