Tools
Tools
Scott, Alexandra Oluwaseun Olaitan (2023) Exploring the Application of Neural Network with Back Propagation in Detecting Fraudulent Transactions within the Ethereum Network. Masters thesis, Dublin, National College of Ireland.
Preview |
PDF (Master of Science)
Download (1MB) | Preview |
Preview |
PDF (Configuration manual)
Download (350kB) | Preview |
Abstract
Blockchain technology, exemplified by the Ethereum network has witnessed widespread adoption in its use. Relying on its robust security features of decentralisation, immutability and cryptography, individuals and establishments have over time gravitated towards cryptocurrencies for financial transactions. However, as with many financial systems, the cryptocurrency domain is not immune to fraudulent activities. With reports of illicit transactions peaking to around USD 14 million in 2021, several researchers have developed robust algorithms all attempting to detect and consequently, prevent the occurrence of these fraudulent transactions.
This study seeks to add to research in this domain by utilising a machine learning algorithm – Neural Network with back propagation in detecting these fraudulent transactions within the Ethereum framework. By leveraging historical Ethereum data from the Kaggle repository which contained known valid and illicit transactions, three supervised machine learning algorithms – Support Vector Machine, K-Nearest Neighbour and XGBoost were built as benchmark models for comparison. This paper goes on to build the neural network with back propagation algorithm and evaluates its performance using three performance metrics – recall, accuracy and specificity.
Overall, this study finds that the three benchmark models had a high performance with regards sensitivity / recall. The proposed model saw a high sensitivity of 0.989 but with mixed results based on other metrics.
Actions (login required)
 |
View Item |