Impact of personal information and reputation system on gig workers’ employment status: an interpretable machine learning-based approach

In the online gig economy marketplace, platforms provide employers with personal information and reputation scores of employed individuals; however, the current research on the gig economy has a single methodology that only allows for a focus on linear relationships, examines a small number of features, and fails to compare the variability of personal confidence and reputation scores. To fill this gap, we utilize the nonlinear relationship data of 11 personal information metrics and 6 reputation system metrics of 20,842 employed individuals on the Freelancer platform, and employ a range of machine learning techniques and interpretable methods to mine the dependencies between 17 metrics and earning power and hiring status. Our findings show that nonlinear models excel in capturing worker traits and job scenarios compared to linear models. Specifically, the XGBoost model exhibited the most adept performance. Moreover, employing interpretable methodologies, we established the importance rankings of various features for distinct tasks, unveiling a general employer inclination toward valuing reputation systems over personal information. Additionally, we unearthed instances of fraudulent ratings within the gig economy platform and identified strategies to mitigate this issue. Lastly, we presented tailored recommendations for platform administrators, employers, and workers.