Fraud Detection on Crowdfunding Platforms Using Multiple Feature Selection Methods

In recent years, crowdfunding has emerged as an alternative funding source for startups and emerging businesses, experiencing significant growth. However, this growth has also led to an increase in fraudulent activities. Despite the potential for fraud in the realm of crowdfunding, there is limited knowledge of the phenomenon due to a lack of data on actual instances of fraudulent campaigns. In this paper, we aim to address this deficiency by collecting and analyzing publicly accessible web and social media data from a hundred fraudulent crowdfunding projects. In order to identify and comprehend the distinguishing characteristics of fraudulent campaigns, we first propose 1) using a wide variety of characteristics of campaign projects and project creators, including their profiles, behavior, social traits, and language; then, 2) we propose to use and combine three well-known multiple feature selection methods, which are based on Correlation-based Feature Selection (CFS), Pearson Correlation Coefficient (PCC), and Information Gain (IG), to identify representative features of fraudulent campaigns. Our approach identifies 10 commonly selected key features of fraudulent crowdfunding campaigns, three of which are new, original findings. We provide and discuss our findings and interpretations on the 10 commonly selected key features in relation to previous studies, based on which we construct a fraud detection model with 82.04% accuracy. We also employ Shapley Additive ExPlanations (SHAP) to interpret the fraud detection model, explaining the importance of each feature.