Hybrid CNN-BiLSTM model with HHO feature selection for enhanced fake news detection

Abstract

Fake news, deliberately incorrect or misleading information presented as news, spreads through diverse platforms like online forums and traditional media. Its potential to harm lives, stir panic, and disrupt democratic systems makes it a pressing issue. Detecting fake news automatically is crucial to mitigate its impact. However, previous research often struggles with feature redundancy, suboptimal feature selection, and inefficient classification models, leading to lower accuracy and poor generalizability across datasets. To overcome these limitations, the present work introduces a methodology consisting of four distinct phases to detect fake news involving data preparation, feature extraction, feature selection, and classification. Data preprocessing eliminates ambiguities from the considered datasets and Linguistic Features (LFs) are extracted during the feature extraction phase to further perform the feature selection using Harris Hawks Optimization (HHO) algorithm. Furthermore, the CNN-BiLSTM hybrid model is then employed for classification, leveraging CNN’s spatial feature extraction and BiLSTM’s sequential learning capabilities for improved accuracy. In this work, we have performed the experiments on four publicly available datasets named ISOT, Kaggle, ConFake, and McIntire, where we achieved an accuracy of 98.89%, 98.25%, 98.56%, and 90.26%, respectively. Remarkably, this approach surpasses state-of-the-art methods by achieving an improved accuracy utilizing the HHO feature selection and CNN-BiLSTM hybrid model for fake news detection.