Institutional Repository of Vilnius University
Binarinio atsako algoritmų taikymas apgaulingų pavedimų atpažinimui /
Abstract
dc:descriptionFinancial crime is moving into the virtual space with the expansion of bank's digital systems. Financial crime causes significant annual losses for banks and their customers. One branch of financial crime is fraudulent payments. The high volume of banking payments does not allow financial crime specialists to screen all of the payments, creating a demand for a fast and well-functioning system, capable of identifying fraudulent payments. A system that can quickly and accurately identify fraudulent payments not only facilitates the work of financial crime specialists, but also prevents the human error factor. The introduction of such a system reduces the risk of financial loss and preserves the bank's reputation. To detect fraudulent payments, K-nearest neighbours, random forest, support vector machine and XGBoost models are constructed and combined to form an ensemble. Fraudulent payments represent a small fraction of the payments, hence models are analyzed not only for unbalanced data, but also by applying oversampling and undersampling methods. The hyperparameters of the models are selected by cross-validation and the performance of the model is assessed by specificity, sensitivity, area under the ROC curve, precision and other metrics. The best metric performance was achieved by the XGBoost model trained on the unbalanced training set, correctly classifying 94.26\% of the fraudulent payments and 93.59\% of the non-fraudulent payments.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2024
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Stabingis, Kristupas,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD210643680&prefLang=en_US