Document Type : Original Article
Authors
1
Faculty of Computer Engineering and Information Technology, Sadjad University, Mashhad, Iran
2
Computer engineering, Faculty of Computer Engineering and Information Technology, Sadjad University of Technology, Mashhad, Iran
10.22034/tjee.2026.68653.5067
Abstract
The rapid spread of fake news on digital platforms has become a major societal challenge, influencing public opinion, political stability, and social trust. Despite extensive research, existing detection models often suffer from limited generalization, weak robustness to noisy or short texts, and reliance on manually engineered features. Moreover, the evolving linguistic and contextual nature of fake news reduces the long-term effectiveness of conventional approaches. To address these limitations, this study proposes a novel hybrid fake news detection framework that, for the first time, integrates Kernel Fuzzy Rough Set (KFRS)–based feature selection with a two-stage ensemble of deep learning and traditional machine learning classifiers. Unlike prior studies that rely solely on either neural or statistical models, the proposed approach combines LSTM and Bi-LSTM networks with Logistic Regression, Support Vector Machine, and XGBoost, leveraging their complementary strengths through soft voting and stacking. The KFRS component plays a critical role in refining feature representations by handling uncertainty and reducing noise, which is particularly beneficial for short and noisy textual data. Skip-Gram word embeddings are employed to capture semantic relationships between words and enhance contextual understanding. The proposed framework is evaluated on three benchmark datasets—LIAR, FakeNewsNet, and FakeEdited—demonstrating consistently strong performance across diverse domains. Ablation experiments further confirm that incorporating KFRS leads to substantial improvements in Recall and F1-score. Overall, the proposed method offers a robust, scalable, and practically deployable solution for real-world fake news detection.
Keywords
)