Software reliability is the probability of failure free operation for specified environment in the given specified time domain. The reliability of software is a crucial and serious characteristic of software. Hence, to estimate, measure, and model the reliability of software many researchers have proposed over 200 software reliability models in the past few decades. But, despite many research efforts to develop software reliability prediction models, there is no single parametric model that is available in all cases efficiently. And also in non-parametric software reliability prediction model, model inconsistency and failure detection with failure correction needs tiny emphasis for reliability prediction. Deep learning approaches have a better capability to make more robust predictions of the model than other optimization algorithms and machine learning. So we intended to employ software detection and correction reliability techniques to make the model more robust using our proposed approach. Thus, in this work, we build a model for automatic prediction of software reliability by using previous software failure reports reported during the testing phase in the software development process. The proposed model encompasses three components: preprocessing, feature learning, and prediction. In data preprocessing, we normalized failure data to standard data between 0.1 and 0.9 to overcome the convergence problem. To extract temporal features and prediction, we employed the merits of a convolutional neural network (CNN) and long short-term memory (LSTM) network. We implement the proposed system using Keras (using TensorFlow as a backend) in Python and tested using a sample historical software failure dataset which we got from software reliability respiratory. Proposed model is conducted with four different dataset. The model achieved a prediction performance of 0.0275 RMSE for training and 0.355 for testing in the prediction of software reliability in terms of failure detection and correction for experiment one. The proposed model shows that the improvement of the performance of state-of-the-art models as well by a minimum of 0.0207 RMSE with LSTM and 26.778 RMSE ANN. Besides, a software reliability model is efficient to predict the software reliability metrics when we have sufficient historical failure data for the software.