A carregar...
Publicação
Using predictive analytics for money mule detection on a cryptocurrency exchange
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 1.88 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Cryptocurrencies and blockchain, the novel technology that became widespread with Bitcoin implementation in 2009, offer many applications. While the new technology facilitates fast and pseudo-anonymous transactions, it also leaves room to be exploited for illicit activities, such as money laundering. This dissertation focuses on developing a predictive analytics model with supervised machine learning algorithms for classifying money mule fraud instances.
Money mules are an instrument for the layering stage of money laundering and are used by criminals to hide the origins of their wealth. As confirmed cases were a rare event, the algorithms used were optimised for imbalanced data in addition to trying resampling techniques for under and oversampling the dataset. The algorithms used were logistic regression, decision trees and ensemble methods – random forest and two types of gradient boosting: XGBoost and LightGBM. The most promising results were achieved with the random forest algorithm, as it reached the best result metrics values aligned with the given business objective. However, the study concluded that the business objective could not be fully realised with the developed model, as it falsely predicts a percentage of the events, which could cause constraints on the business. Therefore, the selected company can use the results and models as a reference tool and a base for further data analyses.
Descrição
Dissertation presented as the partial requirement for obtaining a Master's degree in Information Management, specialization in Information Systems and Technologies Management
Palavras-chave
Cryptocurrency Money Laundering Money Mule Predictive Analytics Machine Learning