Fake news has become a critical societal concern due to its capacity to influence public opinion and compromise the reliability of information. This paper examines advanced approaches to fake news detection, including machine learning models, propagation-based techniques utilizing geometric deep learning, and ensemble methods. It compares various methodologies, features, and performance metrics from leading studies, highlighting the strengths and weaknesses of pre-trained models like BiLSTM, graph-based propagation techniques, and ensemble learning strategies. Through the incorporation of cutting-edge natural language processing (NLP) techniques with classifiers like Naïve Bayes, Random Forest, Logistic Regression, XGboost, Catboost and Support Vector Machines, CNN, BiLSTM, LSTM and their ensemble techniques. The study creates a thorough framework for identifying false information. Performance is assessed through the utilization of dataset such as Kaggle data.csv use metrics like F1-score, recall, accuracy, and precision, demonstrating effective feature engineering and ensemble approaches can enhance detection performance. Ensemble the approaches it gets 98% accuracy. The proposed system facilitates both static and dynamic analysis, enabling real-time classification and validation of sources. Future research directions include the development of hybrid models, multilingual approaches, and early detection systems to address the ongoing challenges of fake news in the modern era.