The threat of fake news jeopardizing the credibility of online news platforms, particularly on social media, underscores the need for innovative solutions. This paper proposes a creative engine for detecting fake news, leveraging advanced machine learning techniques, specifically Bidirectional En-coder Representations by Transformers (BERT). Our approach involves feature selection from news content and social contexts, combining predictions from multiple models, including Random Forest, BERT, GRU, LSTM, and a voting ensemble model. Through extensive evaluation of the WELFake dataset, our method highlights an impressive accuracy of 99%, surpassing baselines and existing systems. Our study highlights the crucial role of hyperparameter tuning, improving the performance of the BERT model to 100%.