Analisis Kinerja Logistic Regression dan Random Forest pada Deteksi Fraud E-Commerce Menggunakan SMOTE dan PCA

Abstract

The rapid growth of e-commerce platforms has increased the volume and complexity of digital transactions, which is accompanied by a rising risk of fraudulent activities. This study aims to apply and evaluate the performance of Logistic Regression and Random Forest algorithms for fraud detection in e-commerce transactions. To address the class imbalance problem, the Synthetic Minority Over-sampling Technique (SMOTE) is employed, while dimensionality reduction is performed using Principal Component Analysis (PCA). The dataset is divided into training and testing sets using an 80:20 ratio. Model evaluation is conducted under four scenarios: baseline without additional preprocessing, SMOTE only, PCA only, and a combination of SMOTE and PCA. The results indicate that Random Forest consistently outperforms Logistic Regression across most evaluation metrics, including Recall, F1-Score, and Area Under the Curve (AUC). The application of SMOTE significantly improves the model’s ability to identify fraudulent transactions, achieving the highest Recall of 80.79% using Random Forest. In contrast, the use of PCA, either alone or combined with SMOTE, tends to degrade model performance. This study concludes that Random Forest combined with SMOTE provides the most effective approach for fraud detection in highly imbalanced e-commerce transaction data.