Internal Core System User Fraud Prediction Using Machine Learning: the Case of Commercial Bank of Ethiopia
| dc.contributor.advisor | Michael Melese | |
| dc.contributor.author | Abayneh Aklilu | |
| dc.date.accessioned | 2025-08-31T22:54:35Z | |
| dc.date.available | 2025-08-31T22:54:35Z | |
| dc.date.issued | 2024-01 | |
| dc.description.abstract | Internal user fraud identification is critical to the operation and growth of any business. Identification of fraudulent users can help businesses understand the reasons for internal fraud and plan the bank's strategies accordingly to boost business growth. Internal core system users, such as CBE employees and authorized personnel with privileged access to core banking systems, pose a risk for internal fraud. This type of fraud involves unauthorized access, misuse of privileges for malicious purposes like data theft or unauthorized transactions, and manipulation of internal users through social engineering. Detecting such fraud can be challenging due to insiders' deep understanding of the systems they exploit. The goal of this research is to develop a machine learning model that can accurately predict internal core system users’ fraud from the Commercial Bank of Ethiopia Internal core system users. For this study, a total of 7,754 datasets with twelve attributes have been used. To determine the best classifier, the model's overall accuracy was used as the evaluation metric. To this end, supervised machine learning techniques Logistic Regression, Random Forest, Support Vector Machine, and K-Nearest Neighbor were applied to predict internal core system user fraud in the situation of the Commercial Bank of Ethiopia. Based on the previous literature, those have been widely used classifier algorithms for fraud prediction. In this study, attribute selection was conducted through the utilization of the correlation matrix and feature importance. The process involved identifying the variables that demonstrate the strongest correlation with the outcome variable and those that have the most significant predictive capability. Furthermore, the selected algorithm's efficiency was evaluated and compared after balancing the data using the SMOTE technique. The best overall classifier is Random Forest (RF) with an accuracy of almost 71.63%, a precision of 96.90%, and recalls also 72.96%; then Logistic Regression (LR), with an accuracy of almost 57.96%, a precision of 97.10%, and recall 58.21%, K-Nearest Neighbor (KNN) with an accuracy of almost 47.52%, a precision of 97.63%, and recall 46.80%, and Support Vector Machine (SVM) with an accuracy of almost 54.74%, a precision of 97.04%, and recall also 54.58% respectively. | |
| dc.identifier.uri | https://etd.aau.edu.et/handle/123456789/7264 | |
| dc.language.iso | en_US | |
| dc.publisher | Addis Ababa University | |
| dc.subject | Internal Core System User Fraud Prediction | |
| dc.subject | Internal Core System User Fraud | |
| dc.subject | Machine Learning | |
| dc.title | Internal Core System User Fraud Prediction Using Machine Learning: the Case of Commercial Bank of Ethiopia | |
| dc.type | Thesis |