Explainable Artificial Intelligence For Pseudo Social Media Profile Detection
Trisha Ghosh, Rupali P. Patil · 2024
Due to the increasing frequency of fake profiles and their detrimental effects, which include spreading rumors and fa-cilitating identity theft, social media platforms like Instagram and Twitter are having a greater and bigger impact on people's lives. This study presents hybrid machine learning models like Deep Neural Network-Support Vector Machine (DNN-SVM), Long Short-Term Memory-Support Vector Machine (LSTM-SVM) and Artificial Neural Networks (ANN) achieving high accuracy that detect fake accounts on social media platforms, ensuring online integrity and security. These models have attained an accuracy of 99% on Twitter dataset, and 96% on Instagram dataset. Along with this, to address the lack of explanation capabilities in existing detection systems, explainable artificial intelligence (XAI) techniques such as SHapley Additive exPlanations (SHAP) values and Local Interpretable Model-agnostic Explanations (LIME) methods are incorporated. In the Instagram dataset, critical features included userFollowerCount, userFollowingCount and user-MediaCount, while statuses_count, favourites_count, geo_enabled and lang_num emerged as influential features in the Twitter dataset. The goal of this study is to detect fake social media accounts and provide transparency and understanding regarding the rationale behind machine learning predictions by elucidating the decision-making processes of the models, aiming to instill confidence in their outcomes and facilitate informed decision-making in real-world applications.