The current work proposes a hybrid deep learning approach for classification of software security requirements based on the DOSSPRE dataset.[12] publish a work on software requirements classification where a software requirement dataset named DOSSPRE have been generated. They mentioned that the non classification of security requirements has been identified as a major source of security concerns in the software development process. For that reason, they give an approach to classify security requirement, but they didn’t consider deep learning as classification technique. The literature review show that, the application of machine learning techniques for software requirements classification is increasing these last years but in the most of application cases classical assemblage techniques are prioritized and deep learning techniques are used in just few cases, due to his complexity and the necessary resources like the needed amount of input data and a high computational power to achieved the expected results. This study contributes to fill the gap in application of deep learning technique for software security requirement classification. A novel approach for hybrid machine learning model is proposed based on naïve bayes model (MNB_Model) and a deep learning model. Three hybrid machine learning models are compared: a deep neural network (DNN-MNB_Model), a convolutional neural network (CNN-MNB_Model) and a recurrent neural network (RNN-MNB_Model). The results of the comparison show that the CNN-MNB_Model get the best performance with 80.2% accuracy.