Spam account detection in online social networks

Online social networks rank first among the most commonly used social tools for spreading information. On social networks, the number of followers is often regarded as an indicator of success, and this number is frequently associated with the accuracy of the news shared. However, by purchasing fake followers through fake accounts, it is possible to increase the follower count, making the account appear legitimate, allowing misinformation to be easily spread. Therefore, in our study, we aimed to make social networks more reliable by identifying these fake accounts and controlling their activities. Our research focused on the Instagram social network. To obtain accuracy results, tree-based learning algorithms such as Random Forest, Extra Trees, Gradient Boosting, Decision Tree, and AdaBoost (Adaptive Boosting) were used. As a result of the necessary data analyses, we observed that these algorithms achieved accuracy rates of 94%, 93%, 90%, 88%, and 94%, respectively. Unlike other studies, to further improve the results of the algorithms, we used ensemble learning methods, specifically the VotingClassifier (hard voting) and StackingClassifier (meta learner) algorithms. The primary reason for using these methods is that combining the advantages of multiple models can provide better performance than a single model. In our study, by combining Random Forest, Extra Trees, and AdaBoost algorithms in the VotingClassifier and StackingClassifier models, we achieved the highest accuracy rates of 96% and 95%, respectively. Another feature that distinguishes our study from others is the use of the YoloV8 object recognition model to categorize and analyze profile pictures through image processing techniques, focusing on the impact of profile pictures on fake accounts. The reason for using the YoloV8 model is that it provides faster and more accurate improvements in results. Finally, the analysis results we obtained were tested for fake account ownership using real-time data pulled from Instagram's API through a Windows interface developed on the Visual Studio platform. As a result, by using both image processing and machine learning techniques together, we have contributed to the literature by increasing the accuracy of the results.