Scholarlink

<p>E-commerce allows for the global buying and selling of goods and services, offering convenience and accessibility to consumers and businesses. However, it is vulnerable to fraud, with malicious transactions posing significant risks. Existing methods do not address changes in user behaviour over time for effective fraud detection. Therefore, this paper proposes a fraud detection method using FSMNN and considers dynamic user behaviour. Firstly, data is collected from the Fraudulent E-commerce Transactions dataset and pre-processed. KDA-BMOT is then used for data balancing. Next, categorical features are extracted and one-hot encoded, followed by outlier detection and removal using ELIPOF for numerical and encoded data. Data is then standardized using log transformation, aggregated temporally using RG-TCMM, and a transaction graph is built using an Adjacency List Graph (ALG). Change points are detected using the Cumulative Quantile Control Chart (CQCC) from the graph data and temporal aggregation. Features are extracted from the aggregated outcome, transaction graph, and detected change points. The FSMNN classifier is employed with transfer learning to detect fraud and normal transactions stored on the cloud. In real-time, IoT devices collect transaction details, which are stored in the cloud, analyzed for fraud detection, and updated accordingly. The proposed FSMNN detected the fraud transactions effectively with an accuracy of 97.58% and an F1-score of 98.95%, respectively.</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

Cloud and IOT data based real-time fraud detection in E-commerce transactions using FSMNN approach

  • 期刊论文

  • 发表日期：

    2025年3月29日

查看全文
  • Ramya Lakshmi Bolla,
  • Rajeswaran Ayyadurai,
  • Karthikeyan Parthasarathy,
  • Naresh Kumar Reddy Panga,
  • Jyothi Bobba,
  • Roseline Oluwaseun Ogundokun

摘要

E-commerce allows for the global buying and selling of goods and services, offering convenience and accessibility to consumers and businesses. However, it is vulnerable to fraud, with malicious transactions posing significant risks. Existing methods do not address changes in user behaviour over time for effective fraud detection. Therefore, this paper proposes a fraud detection method using FSMNN and considers dynamic user behaviour. Firstly, data is collected from the Fraudulent E-commerce Transactions dataset and pre-processed. KDA-BMOT is then used for data balancing. Next, categorical features are extracted and one-hot encoded, followed by outlier detection and removal using ELIPOF for numerical and encoded data. Data is then standardized using log transformation, aggregated temporally using RG-TCMM, and a transaction graph is built using an Adjacency List Graph (ALG). Change points are detected using the Cumulative Quantile Control Chart (CQCC) from the graph data and temporal aggregation. Features are extracted from the aggregated outcome, transaction graph, and detected change points. The FSMNN classifier is employed with transfer learning to detect fraud and normal transactions stored on the cloud. In real-time, IoT devices collect transaction details, which are stored in the cloud, analyzed for fraud detection, and updated accordingly. The proposed FSMNN detected the fraud transactions effectively with an accuracy of 97.58% and an F1-score of 98.95%, respectively.