Klasifikasi Transaksi Penipuan pada Kartu Kredit Menggunakan Arsitektur Multi-Layer Perceptron (MLP)

Abstract

The rapid growth of non-cash payment systems, particularly credit card transactions, has increased convenience in financial activities but also heightened the risk of fraudulent transactions. One of the main challenges in credit card fraud detection is the highly imbalanced nature of transaction data, where fraudulent transactions represent only a small fraction of the total dataset, as well as the complexity of fraud patterns. This study aims to classify credit card fraudulent transactions using a Multi-Layer Perceptron (MLP) architecture as a Deep Neural Network (DNN) approach. The dataset used in this research is the Credit Card Fraud Detection dataset obtained from Kaggle, consisting of 284,807 transactions with anonymized features V1–V28, transaction amount, and class labels. Data preprocessing stages include feature selection, data standardization using StandardScaler, and handling class imbalance through the Synthetic Minority Oversampling Technique (SMOTE). The proposed MLP model is constructed with three hidden layers containing 64, 32, and 16 neurons, ReLU activation functions, a dropout rate of 30%, and a sigmoid activation function in the output layer. The model is trained using the Adam optimizer and Binary Crossentropy loss function, with Early Stopping applied to prevent overfitting. The experimental results show that the MLP model achieves strong classification performance, with a ROC-AUC value exceeding 0.90 and a significant improvement in recall for the fraud class. These results indicate that the proposed MLP-based approach is effective in detecting fraudulent credit card transactions on highly imbalanced datasets. This study is expected to contribute to the development of more accurate and adaptive fraud detection systems in the financial sector.