Fake news has become a critical research area due to the expanding spread of misinformation on digital platforms. As social media becomes a main news source for millions, confirming the authenticity of online material is crucial. This research addresses the problem using a multi-label text classification method, which better identifies the complex nature of misinformation that may fall into multiple categories. In our research, we used the LIAR dataset, a benchmark collection of 12.8k short political statements labeled with six fine-textured truthfulness classes. A transformer-based architecture, notably a fine-tuned BERT model, is proposed including both textual context and metadata to enhance better classification output. The model achieved an accuracy of 0.7351, presenting superior results than those reported earlier research using the LIAR dataset. Confirming the robustness of the approach across different classes with evaluation metrics and per-label precision and recall. The findings illustrate the potency of transformer-based architectures in identifying different degrees of falsehood. This research adds an extensible and practical solution for real-time fake news detection to support information.