NDUBUISI, CHINENYE ROSEMARY (2026) DEVELOPMENT OF BANK ALERT FRAUD DETECTION SYSTEM USING MACHINE LEARNING ALGORITHMS. Other thesis, Godfrey Okoye University, Enugu Nigeria.
![[img]](https://eprints.gouni.edu.ng/style/images/fileicons/text.png) |
Text
ndubuisi-full-project.pdf Download (2MB) |
Abstract
However, the increasing incidence of bank alert fraud, whereby individuals receive fraudulent SMS alerts from unknown senders or even fake banks, poses a potential danger to the use of bank alerting systems. Despite the availability of existing literature on transaction fraud detection using data mining techniques, there are few studies that have been conducted on the classification of user-received SMS alerts. Consequently, this research aims to develop a bank alert fraud detection system for using machine learning algorithms. In this study, a dataset consisting of 600 SMS messages was collected manually. The messages were labeled into three categories: real, fake, and suspicious bank alerts. The dataset was collected manually via gathering of SMS messages from personal acquaintances, family members, and friends, as well as some publicly available SMS messages on social networking websites. The dataset includes 200 samples for each of the classes, after which it is further preprocessed, cleaned, and enlarged to 2,700 SMS messages. This feature set was extracted via TF-IDF Vectorization and Custom Feature Extraction. Three different models, including logistic regression, random forest, and XGBoost, were trained and tested with the data. The performance of the Logistic Regression classifier is better than that of others. It has an accuracy of 97.6%, precision of 97.7%, recall of 97.6%, and F1-score of 97.6%. The application is capable of detecting bank SMS alert fraud. The front-end was made using React Native, and the back-end was coded in FastAPI. Supabase will be used for the database management system. The user can decide to type in his/her message for verification. These messages can be classified either as 'real', 'fake', or 'suspicious'. If users feel like the classification algorithm used in this study has given a false label, they can provide feedback through API points in order to make the classifier better. This study does not need any SMS interception feature since it violates the terms set by both Android and the Google Play Store.
| Item Type: | Thesis (Other) |
|---|---|
| Subjects: | Q Science > Q Science (General) |
| Divisions: | Faculty of Computing And Information Technology (FACIT) |
| Depositing User: | Cynthia Ugwuoti |
| Date Deposited: | 24 Jul 2026 14:57 |
| Last Modified: | 24 Jul 2026 14:57 |
| URI: | http://eprints.gouni.edu.ng/id/eprint/6014 |
Actions (login required)
 |
View Item |
