Scholarlink

<p>The surge in credit card transactions has exacerbated the issue of fraud significantly. According to the 2021 Nilson report, global fraud increased to $32.2 billion and is forecasted to reach $49.32 billion by 2030, underscoring the urgency for effective detection methodologies. The artificial neural network (ANN) is a highly efficient technique for supervised credit card fraud detection among machine learning algorithms. Its adeptness in handling intricate computations on extensive datasets while demonstrating robust generalization capabilities positions it as a powerful tool. This study introduces a pioneering approach that unites two architectures: The Autoencoder (AE) and the Multilayer Perceptron, forming a cohesive Fully Connected Deep Network (FCDN). Leveraging the AE to extract pertinent features from input transactions reduces data complexity while retaining essential information. This dimensionality reduction strategy significantly augments the performance of the Fully Connected Deep Network (FCDN) in detecting fraud. Our experiments were conducted on four imbalanced credit datasets and compared against recent studies. The model showcased a remarkable ability to minimize false positives and negatives, exhibiting superior evaluation metrics encompassing accuracy, recall, precision, and F1 score. Hence, our approach presents a robust solution to combat credit card fraud, bolster security measures, and mitigate financial losses for businesses and individuals.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep neural network for detection of fraudulent transaction

  • 期刊论文

  • 发表日期：

    2025年4月10日

查看全文
  • Fatima Zohra El Hlouli,
  • Jamal Riffi,
  • Mohamed Adnane Mahraz,
  • Ali Yahyaouy,
  • Khalid EL Fazazy,
  • Hamid Tairi

摘要

The surge in credit card transactions has exacerbated the issue of fraud significantly. According to the 2021 Nilson report, global fraud increased to $32.2 billion and is forecasted to reach $49.32 billion by 2030, underscoring the urgency for effective detection methodologies. The artificial neural network (ANN) is a highly efficient technique for supervised credit card fraud detection among machine learning algorithms. Its adeptness in handling intricate computations on extensive datasets while demonstrating robust generalization capabilities positions it as a powerful tool. This study introduces a pioneering approach that unites two architectures: The Autoencoder (AE) and the Multilayer Perceptron, forming a cohesive Fully Connected Deep Network (FCDN). Leveraging the AE to extract pertinent features from input transactions reduces data complexity while retaining essential information. This dimensionality reduction strategy significantly augments the performance of the Fully Connected Deep Network (FCDN) in detecting fraud. Our experiments were conducted on four imbalanced credit datasets and compared against recent studies. The model showcased a remarkable ability to minimize false positives and negatives, exhibiting superior evaluation metrics encompassing accuracy, recall, precision, and F1 score. Hence, our approach presents a robust solution to combat credit card fraud, bolster security measures, and mitigate financial losses for businesses and individuals.