Individual Versus Ensemble Models in Instagram Fraud Accounts Detection Using Machine Learning Models

Abstract

Social media has emerged as a primary platform for the sharing of information and staying in touch. However, the rapid growth of social media has also led to the emergence of fake accounts. These accounts are used for the distribution of false information, the execution of scams and the impersonation of genuine users. Hence, there is a real problem to the security and trust on such platforms. In order to address this issue, this study proposes a machine-learning based method for detecting fake Instagram profiles on the basis of various metadata properties. These properties include the number of followers and followings, Biography, Posting behavior represented by the number of posts and properties of the usernames. Various categorization techniques, like logistic regression, svm, random forest, and catboost, are used in this work. Additionally, a fusion method is proposed which collaborates the results of random forest and CatBoost classification techniques using the average probability value. According to the experimental results, it has been observed that the proposed fusion method has better results compared to the other machine learning techniques.