Dual-Stage Attention with Bi-LSTM and Adaptive XGBOOST for Robust Fake News Detection

The rapid spread of fake news across digital platforms has raised significant concerns regarding misinformation and its societal impact. Traditional fake news detection models rely on deep learning (DL) or ensemble-based approaches but often lack robustness and interpretability. This paper proposes a novel Dual-Stage Attention with Bi-LSTM and Adaptive XGBoost (DAE-BiLSTM-XGB) model for enhanced fake news classification. The first attention stage refines text embeddings by emphasizing contextually significant words, while the second attention stage in the Bi-LSTM module captures long-term dependencies in sequential data. Features extracted through Bi-LSTM are further optimized using an Adaptive XGBoost classifier, improving the system's generalization capacity. Experiments conducted using standard datasets illustrate that DAE-BiLSTM-XGB outperforms existing models in terms of accuracy, F1-score, and robustness against adversarial perturbations. This hybrid approach presents a scalable and explainable solution for real-time misinformation detection in social media and news platforms.