Advancements in Fake News Detection: A Review of Machine Learning and Deep Learning Approaches

The rapid proliferation of fake news on digital platforms has emerged as a global challenge, influencing elections, shaping public perception, and undermining trust in media. The review synthesizes 40 peer-reviewed studies on fake news detection (FND) published between 2017 and 2025. Studies were selected through a systematic search of IEEE Xplore, ACM Digital Library, Scopus, and SpringerLink using pre-defined keywords. Early research focused on traditional ML techniques and propagation-based models, while recent advances incorporate hybrid architectures, transformers, graph neural networks, and multimodal frameworks. The review is organized thematically into six categories: graph- and propagation-aware models, transformer- and embedding-based classifiers, hybrid and ensemble approaches, traditional ML and multilingual methods, comparative and explainable frameworks, and dataset and taxonomy developments. Key findings highlight improvements in scalability, semantic understanding, and contextual modeling, with state-of-the-art accuracies exceeding 95-99% on benchmark datasets such as LIAR, FakeNewsNet, and GossipCop. Despite these advances, persistent challenges remain, including multilingual generalizability, dataset bias, real-time adaptability, and vulnerabilities to AI-generated misinformation. This review contributes by identifying research gaps and proposing future directions, including multilingual benchmarks, explainable AI, vision-language transformers, and privacy-preserving detection methods.