International Journal of Science, Engineering and TechnologyAuthors: Dr. Pankaj Malik, Tanvay Soni, Tanishq Sharma, Rashi Dongre, Sarthak Shrimali
Abstract: The rapid expansion of e-commerce platforms has introduced significant challenges in ensuring the authenticity of vendors and the integrity of supply chains. Traditional fraud detection techniques often fail to capture the complex, dynamic relationships among vendors, products, and transactions. In this study, we propose a novel graph-based machine learning framework that leverages graph embeddings to detect fake vendors in e-commerce supply networks. By modeling the supply ecosystem as a heterogeneous graph comprising vendors, products, transactions, and reviews, we employ node embedding techniques such as Node2Vec and GraphSAGE to learn low-dimensional representations of entities. These embeddings are then fed into supervised classifiers (e.g., Random Forest, XGBoost, and GCN) to identify fraudulent vendors. A labeled dataset was constructed using transaction logs and platform moderation records from a leading e-commerce platform, consisting of 12,000 vendors (1,500 labeled as fake). Our approach achieved a detection accuracy of 94.3%, with a precision of 91.8%, recall of 89.6%, and F1-score of 90.7%, outperforming baseline methods such as rule-based heuristics and traditional feature-based classifiers. Furthermore, the embedding visualizations revealed distinct clusters of suspicious vendor behavior, highlighting the interpretability of our model. The results demonstrate the effectiveness of graph embedding techniques in capturing relational patterns and structural anomalies, offering a scalable and intelligent solution for fraud detection in e-commerce supply chains.
DOI: http://doi.org/10.5281/zenodo.15737544
