Nowadays, social media platform has enhanced tremendously in the last few years and considered one of major source of for information seeking and sharing in low cost. Due to the growth of data, The Fake news are quickly dominating and spreading the information, and distorting the community for sharing their own thoughts, knowledge and point of view regarding towards to any topic. In this paper, structural features with updated RNN and LSTM methods are proposed to improve the performance of system on fake news data. The system uses attention layer with RNN and LSTM to update the weights and values of different features. The performance of model also compared with various hyper parameters such activation, optimization, and dropout. The Proposed Model based with Long Short Term Memory (LSTM) categorize the features closer on original and fake news with customized hyper parameters and random search. The experimental results also depicted that deep learning methods outperformed when size of data samples is high. Furthermore, we showed that combining strong feature engineering with deep learning models, we can more concisely identify the fake news with state-of-art results.