@article{article_1688563, title={Enhancing Fake News Detection: A Multi-Modal Approach Integrating Reinforcement Learning, Random Forest, and LSTM}, journal={Turkish Journal of Engineering}, volume={9}, pages={831–847}, year={2025}, DOI={10.31127/tuje.1688563}, author={Kumar, Vivek and Singh, Partap and Ahmad, Waseem}, keywords={Machine Learning, Deep Learning, Reinforcement LearningLong, Short Term Memory, Fake News}, abstract={Previously, people were not aware of fake news and they thought that all the news coming from social media platforms like YouTube, Twitter, Facebook, online ads, etc. was true. However, when they feel that intentionally spreading such news will damage their business, political, and social reputation, the impact occurs, i.e., they lose interpersonal relationships or monetary reputation in society. Researchers believe that it is very important to detect and understand the truth or falsehood of news so that we can stop the spread of fake news in society. To detect whether the news is fake or not, researchers apply machine learning (ML) methods, and these ML algorithms are increasingly improving the efficiency of fake news detection. This paper uses a hybrid model consisting of reinforcement learning, machine learning (ML), and deep learning (DL) algorithms to detect fake news. Reinforcement learning (RL) improves the feature selection process, random forest (RF) algorithm provides persistent classification, and long short-term memory (LSTM) captures and understands the continuous dependencies in the data. Our hybrid method framework develops fake news detection in a better way. In addition, it provides an understanding of how to combine various effective models to achieve good performance in detecting fake news in the real world. The model is evaluated using three publicly available datasets: FakeNewsNet, COVID-19 fake news, and Kaggle fake news datasets, which are selected for their diversity and relevance. Preprocessing steps include text normalization, tokenization, and lemmatization. The model is trained directly on the original dataset distribution and its performance is monitored to ensure that no bias towards any class occurs. The experimental results show that our hybrid model outperforms individual and traditional ML methods, it achieves an accuracy of 80.6% on FakeNewsNet, 92.6% on the KaggleFakeNews dataset, and 92.9% on the COVID-19 dataset. F1-scores ranged from 81.2% to 92.5%, reflecting balanced performance across precision and recall. Our proposed model achieves state-of-the-art performance compared to other machine learning methods in terms of better performance metrics such as F1 score, accuracy, recall, and precision.}, number={4}, publisher={Murat YAKAR}