Published: Dec. 28, 2023

The proliferation of AI-powered bots and sophisticated fraudsters poses a significant threat to the integrity scientific studies reliant on online surveys across diverse disciplines, including health, social, environmental political sciences. We found substantial decline in usable responses from 75% 10% recent years due survey fraud. Monetary incentives attract capable mimicking genuine open-ended verifying information submitted months prior, showcasing advanced capabilities fraud today. This study evaluates efficacy 31 indicators 6 ensembles using two agriculture California. To evaluate performance each indicator, we use predictive power recall. Predictive is novel variation precision introduced this study, both are simple metrics that allow for non-academic practitioners replicate our methods. best included email address score, MinFraud Risk Score, consecutive submissions, opting-out incentives, improbable location, start time. Despite multiple methodological innovations, none or ensemble tests proved adequate large proportion fraudulent original data samples. Findings underscore evolving tactics fraudsters, demonstrating their increased proficiency responding matching, domain knowledge, questions. conclude with recommendations developing adaptable detection strategies.

Language: Английский