Supervised Machine Learning Algorithms to Detect Instagram Fake Accounts

Abstract

Instagram is extremely popular because many celebrities and their fan pages use Instagram as the platform for them to communicate. Instagram offers many media sharing features and has proven to be the most popular social media platform for promoting many brands. As the most popular platform, Instagram also has fake users. Regrettably, some people do malicious activities using fake accounts such as impersonating artists or influencers, hate comments and spread rumors to become viral. Hence, this research aims to detect Instagram fake users based on the user's profile. There are several stages before account authenticity detection is successful, starting from data pre-processing, selecting a classification model, and classification evaluation. The algorithms that are used to create the supervised machine learning model are Logistic Regression, Bernoulli Naive Bayes, Random Forest, Support Vector Machine, and Artificial Neural Network (ANN). This paper tried two experiments. The first is that the default state of the model has no parameters, and no features are added. Second, to improve the accuracy, new features and tuning parameters were added in the experiment. Models that perform better than other models based on the second experiment with new features and parameters are Logistic Regression and Random Forest, with an accuracy of 0.93.