Design of an efficient model for fake profile detection on social media using advanced feature engineering and deep learning techniques

Detecting false social media profiles is crucial for internet trust and honesty. Existing approaches generally struggle to capture social network data’s complex patterns. New feature engineering, ensemble learning, deep learning architectures, and unsupervised anomaly detection methods are presented in this study to answer the urgent demand for strong fake profile identification. Traditional phony profile detection methods use simple feature engineering, making it difficult to extract relevant information from complicated social network data samples. Deep Feature Synthesis (DFS) automatically develops complicated features to improve the model’s capacity to distinguish bogus profiles from real ones in varied use-case scenarios. Our ensemble learning method, Extreme Gradient Boosting with Regularization and Early Stopping (XGBoost+), outperforms others by using regularization and early stopping criteria. This optimises model performance and prevents overfitting, improving classification accuracy. We also introduce the Graph Attention Network with Temporal Embeddings (GAT-TE), a cutting-edge deep learning architecture that captures social network spatial and temporal dynamics. Self-attention mechanisms help GAT-TE model temporal dependencies better than RNN-based models, improving false profile detection accuracy. Our unsupervised anomaly detection method, Variational Autoencoder with Wasserstein Autoencoder Regularization (VAE-WAE), is innovative for social network data. VAE-WAE detects bogus profiles better by learning robust data representations and avoiding false alarms. Our work affects social media platform integrity, cybersecurity, and online trust. Our research advances bogus profile detection algorithms, making online environments safer and more trustworthy, boosting trust and authenticity in digital interactions