نوع مقاله : مقاله پژوهشی
نویسندگان
1 دانشیار، گروه مدیریت، دانشگاه قم، قم، ایران
2 کارشناسی ارشد، گروه مدیریت فناوری اطلاعات، واحد علوم و تحقیقات، دانشگاه آزاد اسلامی، تهران، ایران
3 گروه مدیریت، دانشکده علوم اقتصادی و اداری، دانشگاه قم، قم، ایران
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
Objective: With the rapid expansion of e-commerce and the growing reliance of users on online shopping, the threat of fraudulent websites has significantly increased. These websites imitate reputable brands, replicate visual designs, use domain names similar to legitimate ones, and create deceptive user experiences to trick users into entering sensitive information such as usernames, passwords, credit card details, verification codes, and personal identification data. The consequences of such attacks are not limited to financial losses; they also damage public trust, harm brand reputation, increase complaint-handling costs, and may even disrupt digital service supply chains. Although numerous studies have examined the application of machine learning and deep learning algorithms in fraud detection, a major gap in the literature remains the absence of a practical, prioritized, and decision-support framework. Such a framework should guide organizations in selecting which capabilities and indicators to implement under real-world constraints, including limited budgets and human resources. Most previous studies primarily focus on improving model accuracy while overlooking practical feasibility, economic benefits, and implementation and maintenance costs in an integrated manner. The objective of this research is to fill this gap by developing a systematic framework for identifying, evaluating, and prioritizing machine learning capabilities for detecting fraudulent e-commerce websites. This framework enables decision-makers to adopt a balanced perspective that considers both technical and managerial criteria when planning implementation strategies.
Methodology: This study is applied and quantitative in nature. In the first stage, a systematic review of scientific and industry sources (over 150 articles and reports) was conducted to extract relevant capabilities for detecting fraudulent websites. As a result, 25 capabilities were identified, covering various dimensions: domain and URL features (e.g., domain length, lexical patterns, structural similarity), website content (text, images, keywords, and page structure), user behavioral patterns (navigation behavior, click patterns, dwell time), financial transaction data (payment anomalies and suspicious patterns), and web page changes over time (abnormal and sudden updates). Next, the fuzzy Delphi method was employed to assess the importance and usability of each capability from experts’ perspectives, allowing for the management of uncertainty and differences in judgment. In this phase, nine capabilities with a defuzzified score above 0.7 advanced to the final stage. Subsequently, the multi-criteria decision-making method (MARCOS) was applied to prioritize the selected capabilities based on three criteria: feasibility, financial benefits, and implementation cost. Expert data were aggregated, normalized, and weighted to determine the final ranking of each capability.
Findings: The results indicate that five capabilities have the highest priority and impact in detecting fraudulent e-commerce websites: (1) Classifying websites as fraudulent or legitimate using machine learning and deep learning algorithms; (2) Analyzing suspicious financial transactions and identifying abnormal payment flow patterns; (3) Detecting rapid and unusual changes in web pages and website content; (4) Identifying suspicious trends and patterns in historical data and user behavior using predictive and pattern-based learning techniques; and (5) Detecting fake domains and domains similar to legitimate ones through URL feature analysis and structural similarity assessment. Other selected capabilities also contribute to improving detection accuracy and reducing risk; however, they rank lower in terms of implementation priority and overall impact.
Conclusion: By providing a prioritized decision-support framework, this study bridges the gap between algorithm-focused research and the practical needs of organizations. The findings demonstrate that a combined approach integrating domain analysis, content analysis, user behavior monitoring, and financial transaction evaluation achieves higher effectiveness in the timely detection of fraudulent websites. The proposed framework can serve as a foundation for designing, investing in, and deploying digital security systems within e-commerce platforms. The study is limited by its reliance on secondary data and a relatively small number of experts. Future research is recommended to utilize real-world datasets, larger expert samples, field validation, and real-time analytical approaches to enhance accuracy and generalizability.
کلیدواژهها [English]
)
ارسال نظر در مورد این مقاله