Akademik Veri Yönetim Sistemi
ACTA POLYTECHNICA HUNGARICA, cilt.23, sa.2, ss.263-282, 2026 (SCI-Expanded, Scopus)
The rapid growth in e-commerce and online payment systems has led to an increase in cyberfraud activities and serious financial losses for commercial enterprises. Solving this critical problem requires the development of effective fraud prevention mechanisms and the investment of online shopping platforms in cybersecurity systems. Fraud detection is usually based on binary classification methods, but reducing the size of high-dimensional datasets is of great importance to detect fraudulent activities with high accuracy. This study proposes a two-stage hybrid framework to prevent credit card fraud. In the first stage, the dimensionality in the dataset is reduced using Autoencoder (AE), and the low-dimensional data obtained from the fully connected layer of the Autoencoder is presented as input data to the Random Forest (RF) classifier in the second stage. The proposed AE+RF model is compared with popular models such as Artificial Neural Networks (ANN), Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN) and XgBoost on the basic performance metrics of precision, recall, F1-Score, AUC-ROC and AUC-PR. Experimental results show that the AE+RF model outperforms its closest competitors by 2.11% in recall, 1.03% in F1-Score, 1.02% in AUC-ROC and 2.08% in AUC-PR. These findings reveal that the proposed framework makes a significant contribution to enhancing the security of e-commerce platforms by providing high accuracy and efficiency in cybersecurity fraud detection.