The Software Quality Prediction Model Based on DBN

Abstract

Although the BP neural network is recognized as an advanced method for predicting software quality, it still has some drawbacks, which is easy to fall into local minimum solution when training the data with complex structure. In this condition, the predict results are inconsistent with the actual software reliability. Deep learning is an improvement and extension for the neural networks, and deep belief networks (DBN) have a network model of deep structure. As a kind of deep learning model, deep belief networks (DBN) is not easy to fall into local minimum solution and has a higher Match Ratio than BP neural network than BP neural network when training the data with complex structure. Due to these good properties, we try to use DBN model to predict software safety based on the data of software code features, and it has been proved that, compared with BP neural network, DBN model is a better model to predict the software quality.