When Simpler Models Outperform Hybrid Ensemble: Re-evaluating Hybrid Ensembles for Financial Fraud Detection

Bajaj, Sanidhya (2025) When Simpler Models Outperform Hybrid Ensemble: Re-evaluating Hybrid Ensembles for Financial Fraud Detection. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (760kB) | Preview
Preview	PDF (Configuration Manual) Download (4MB) | Preview

Abstract

Fraud detection in the financial services sector is important as the consequences of a mere error can result in huge losses. This research runs counter to the popular belief according to which hybrid models, like voting and stacking are always better than individual machine learning models. We assessed this with the help of a comparative assessment of simple machine learning and hybrid techniques with distinctive ensemble. The research conducted used transactional dataset that contained 50,000 records of transaction labelled to indicate that the transaction occurred was legitimate or fraudulent.

Among the hybrid ensemble methods that were evaluated were voting and stacking classifiers compared to simpler but high-accuracy individual classifiers such as Random Forest, Extra Trees and AdaBoost etc. Measurements of assessment were centered towards F1-score, recall and AUC-ROC of classification regarding the fraud class in particular because of the class-imbalanced nature of the dataset. The results showed that even with respect to detection of instances of fraud among the minorities, the more simplistic models provided at least the same and, on most occasions, better results than the hybrid ensemble models. These findings lead to the suggestion that the default use of complex ensemble procedures might not always be the best solution in investigating fraud detection but rather it is utmost to select and optimize models carefully.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Subhnil, Shubham UNSPECIFIED
Subjects:	H Social Sciences > HG Finance H Social Sciences > HG Finance > Financial Services Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	24 Aug 2026 15:43
Last Modified:	24 Aug 2026 15:43
URI:	https://norma.ncirl.ie/id/eprint/9618

Actions (login required)

View Item