| 摘要: | 隨著電子商務的蓬勃發展，信用卡支付已成為線上交易中不可或缺的重要環節。然而，相伴而生的信用卡欺詐行為亦日趨猖獗；對消費者權益、商家營收以及金融體系造成了嚴重衝擊。由於欺詐手法層出不窮、變化多端，傳統基於規則的風險驗證機制已很難全面識別並防範這些新型欺詐威脅。有鑑於此，構建高效精準的信用卡欺詐檢測模型以維護電商交易安全迫在眉睫。
本研究以台灣B2C電子商務購物平台網站為研究個案，目的是透過機器學習技術完善既有的RBRA(Rule-Based Risk Authentication)信用卡偽冒防欺詐框架。本研究收集並預處理了個案研究B2C電子商務購物平台網站2022年的信用卡交易記錄。並分別基於決策樹、隨機森林、邏輯回歸等多種機器學習算法訓練分類模型，以預測每一筆交易的潛在欺詐風險程度。且鑑於本個案研究樣本中信用卡偽冒詐欺類別資料不平衡的特性，本研究採用了SMOTE技術進行資料集增強，引入F1-Score、ROC曲線等指標作為模型評估的主要依據。並由混淆矩陣分析結果進一步印證實驗模型的分類精準度。此外，本研究也探討了信用卡偽冒欺詐交易時間熱點、高風險商品類別以及收件人地區等維度的特徵資料選取。
在理論與實踐相結合的基礎上，本研究提出將優化後的機器學習模型整合電子商務購物平台網站，實現每一筆信用卡交易的即時風險評估。對於高風險訂單，結帳流程將導向到發卡行執行3DS身份驗證，從而形成前後台協同的防欺詐架構，有效維護交易安全性、提升消費者使用體驗。總結來看，本研究為電子商務購物平台網站防範金融欺詐活動提供了技術路徑，對於維護電子商務健康發展和消費者權益具有重要意義。
關鍵字：#機器學習 #資料探勘 #信用卡詐欺偵測 #風險預測 #RBRA
;The surge in credit card fraud accompanying e-commerce growth poses severe threats, necessitating efficient fraud detection models to safeguard transaction security. This study employed a Taiwanese e-commerce platform to enhance the existing Rule-Based Risk Authentication (RBRA) framework using machine learning techniques.
We collected and preprocessed 2022 transaction data, including confirmed fraud cases, and trained classification models using algorithms like decision trees, random forests, and logistic regression. Addressing class imbalance with oversampling, experimental results demonstrated the random forest model′s superior performance, achieving an AUC of 0.976 and F1-score of 0.916, corroborated by exceptional classification accuracy. Feature analysis provided insights into fraud hotspots, high-risk product categories, and recipients′ geographic patterns, informing targeted prevention strategies. Integrating the optimized model enables real-time risk assessment, redirecting high-risk orders for 3D Secure verification while maintaining security and user experience. This study offers a technical pathway for e-commerce platforms to combat fraud, contributing to sustainable development and consumer protection. Leveraging machine learning optimizes fraud detection, fostering a secure online shopping environment.
Future work focuses on incorporating diverse data sources, exploring ensemble models, and enhancing maintainability to further improve real-time performance, accuracy, and robustness, cultivating an efficient and secure e-commerce ecosystem.
Keywords: #Machine Learning #Data Mining #Credit Card Fraud Detection #Risk Prediction #RBRA |
