The conventional algorithm can not deal with the deviant attitude of different experts in the evaluation of information security. At the same time, it adopts shallow machine learning models and mishandled the deviation accumulating problem. In the study, we proposed a novel evaluation algorithm of information security based on the fuzzy adjustment feature matrices and the deep time sequence model. In the proposed algorithm, the triangle fuzzy function was applied to build the expert’s evaluation indexes. Then, the improved weight DS theory of evidence was used to adjust the indexes. After that, the loss matrix and possibility matrix are constructed. The information security was evaluated by a deep neural model in proposed model. In the simulations of MIT dataset, we explored whether the feature matrices will cope with the conflicts of experts or not, and the accuracy, performance and robustness of estimation were also checked. It can be found that the proposed algorithm has a better ability to evaluate the fuzzy system, and the ability for coping with the conflicts of experts is also better, achieving a better accuracy and robustness with the reservation of performance.