A Voting Based Ensemble Approach for Fake Account Detection using Multimodal Social Media Data

The proliferation of online networks has contributed to a growing concern regarding the surge of fraudulent user profiles, which undermine online security and digital credibility. This study introduces a robust framework for identifying fake accounts by leveraging multimodal features derived from both textual content and numerical metadata. Initially, three deep learning architectures, Artificial Neural Network (ANN), Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM) networks, were developed and assessed for their classification capabilities. To improve detection performance, a Voting Classifier was employed, integrating XGBoost, Random Forest, and Gaussian Naive Bayes algorithms. The comparative results indicated that the ensemble model achieved superior performance across key evaluation metrics, including accuracy, precision, recall, and F1-score. By harnessing the complementary strengths of multiple models, the proposed method delivers a dependable solution for identifying deceptive accounts. This research contributes to enhancing the effectiveness of automated fake account detection and encourages further exploration of hybrid models using multimodal inputs.