Abstract
In recent years, the widespread dissemination of fake news on social media has raised concerns about its impact on public opinion, trust, and decision-making. Addressing the limitations of traditional detection methods, this study introduces a hybrid deep learning approach that enhances the identification of fake news. The objective is to improve detection accuracy and model robustness by combining a Long Short-Term Memory (LSTM) network for contextual feature extraction with a Convolutional Gaussian Perceptron Neural Network (CGPNN) for classification. To further optimize performance, we integrated a metaheuristic Moth-Flame Whale Optimization (MFWO) algorithm for hyperparameter tuning. Experimental evaluation was conducted on four benchmark datasets ISOT, Fakeddit, BuzzFeedNews, and FakeNewsNet using standardized preprocessing techniques and TF-IDF-based text representation. Results show that the proposed model outperforms existing methods, achieving up to 98% accuracy, 95% F1-score, and statistically significant improvements (p < 0.05) over transformer-based and graph neural network models. These findings suggest that the hybrid framework effectively captures linguistic patterns and textual irregularities in deceptive content. The proposed method offers a scalable and efficient solution for fake news detection with practical applications in social media monitoring, digital journalism, and public awareness campaigns. Overall, the framework delivers 3-8% higher accuracy and F1-score compared to state-of-the-art approaches, demonstrating both robustness and practical applicability for large-scale fake news detection.