Author(s) :

Nikita Efimov.

Volume/Issue :

Abstract :

The existence of fake users on social media platforms like Instagram creates significant challenges for marketers and platform integrity. Fake users are usually used for engagement manipulation such as spamming and fake followings. This study investigates the problem of identifying fake users using machine learning techniques, leveraging rich metadata from a dataset with 65326 Instagram accounts. Using analysis of 18 metadata features, two models -Random Forest and XGBoost-were trained and evaluated using precision, recall and F1-score. XGBoost achieved the best performance, with F1-score of 0.91 for real users and 0.9 for fake users. Feature importance analysis using SHAP, underlined link availability, engagement rate (comments), and engagement rate (likes) as the most predictive features. This study underscores the potential of machine learning in combating fake user expansion and provides insights for improving model interpretability and efficiency. Future research could explore larger datasets, integrate more advanced techniques and incorporate additional metadata to further refine detection models.