Machine Learning Approaches for Trust Prediction in Online Auction Platforms: A Comprehensive Review

Abstract

Online auction platforms have grown enormously over the past two decades, yet trust between participants remains a serious problem. Fraud, shill bidding, and misleading seller behavior keep undermining buyer confidence, and the need for reliable, automated ways to assess trustworthiness has never been greater. This review looks closely at the use of machine learning (ML) methods for predicting trust in online auction settings, focusing on research published from 2005 to 2024. This study covers the full methodological spectrum: classical approaches like logistic regression, support vector machines, and random forests, through to modern deep learning architectures including recurrent networks, graph neural networks, and transformer models. This paper also examined hybrid frameworks that weave together behavioral analytics, NLP-based review analysis, and social network signals. The benchmark datasets the community relies on are discussed, how performance is measured, and the recurring challenges researchers keep running into, such as class imbalance, feature engineering bottlenecks, concept drift, and the cold-start problem are highlighted. The comparative analysis shows that graph-based models using social trust propagation and ensemble methods that combine various feature sets often perform best. Many achieve macro F1 scores above 0.92 on recognized benchmarks. The paper ends with a research roadmap that highlights open problems and new opportunities. These include using federated learning to maintain privacy during trust assessments, integrating large language models, and developing real-time adaptive trust scoring. This work provides a solid reference for researchers and practitioners in machine learning, e-commerce security, and computational trust.