Crowdfunding Fraud Detection: A Systematic Review Highlights AI and Blockchain using Topic Modeling

Abstract

Crowdfunding platforms have gained popularity as a means of financing entrepreneurial initiatives but face a high risk of fraud. Fraud is a significant problem due to its impact on trust, ultimately leading to financial instability. Detecting and preventing fraud is therefore paramount for the sustainability of crowdfunding platforms. This study provides a systematic review of the literature and state-of-the-art discussions about crowdfunding fraud. Unsupervised topic modeling highlights that both AI and blockchain are recurrently presented in the literature as effective methodologies for identifying and preventing fraudulent practices. Furthermore, this work describes current market practices of crowdfunding platforms in preventing fraudulent behavior and argues that, while fraud is rare, its high impact necessitates new and innovative forms of fraud detection. A key limiting factor for the application of AI solutions is the lack of available labeled crowdfunding data for training efficient algorithms for fraud detection, which is crucial as it constitutes an anomaly detection machine learning task. In this context, unsupervised machine learning methods are discussed as valuable techniques for detecting anomalies in the absence of labeled fraud cases due to their ability to adapt to evolving fraud patterns. Altogether, this research provides valuable insights into the complexity of detecting and preventing fraudulent activities in crowdfunding and highlights effective detection techniques that, if implemented, offer promising solutions to enhance platform reputation and ensure regulatory compliance.