THE USE OF SUPERVISED MACHINE LEARNING CLASSIFIERS FOR THE DETECTION OF FAKE INSTAGRAM ACCOUNTS

Abstract

There has been a meteoric rise in global internet usage since the creation of the internet. Cybercrime and the prevalence of online con artists have both increased as a result. Recent years have seen an increase in the number of reported incidents of online scams, which is likely attributable in part to the widespread use of social media, which serves as one of the most prominent forums for fraudsters to target their victims. Scammers mostly target users of social media platforms like Instagram, Facebook, WhatsApp, and Twitter since these networks facilitate their illegal activities. Instagram's meteoric rise to fame can be directly attributed to the sheer quantity of celebrities and fan pages who frequent the service. As a result of its user base's ability to easily share a wide variety of media, Instagram has quickly become the social media platform of choice for many companies. Instagram may be the most popular social media platform, however, it has been found to feature photographs of fictitious individuals. Unfortunately, some people impersonate artists or influencers, write rude comments, and propagate rumours by creating fake accounts in an attempt to get their content shared widely and gain attention. This is done to draw attention to their own material. Therefore, the purpose of this research is to identify, based on the profiles of the people in question, how to identify fraudulent Instagram users. Multiple steps, including data preprocessing, model selection, and assessment of classification performance, must be completed before a genuine account can be identified with any degree of certainty. The supervised machine learning model is developed using three distinct algorithms: logistic regression, k-nearest neighbour, and a decision tree. This investigation included two distinct experiments. For starters, the model is useless in its current state because it has no parameters and no features. New additions and fine-tuning adjustments were made to the experiment to improve its accuracy. Based on the results of the second experiment, it was discovered which models outperformed the others by a significant margin. If we compare the accuracy of the three models, the decision tree comes out on top with a 96% success rate.