A Novel Approach of Software Fault Prediction Using Deep Learning Technique

Abstract

Now-a-days, failure of the software is unavoidable due to increasing size and complexity of software. So, fault finding is necessary for removing the software faults. Spectrum-based fault localization is most popular technique to find the faulty statements of a given program. Still, there are some limitations also. In case of large software, it is very hard and time taking to test all possible scenarios via traditional approach. The machine learning model is an interesting approach for solving this. Recently, deep learning is widely used for improving the fault finding techniques. Deep learning models are based on the architecture of neural network. The neural network architectures based on input layer, hidden layer(s) and output layer. Convolution Neural Network (CNN) is a well-known architecture for deep learning. The network is trained with large amounts of data and neural network architectures that learn the features directly from the information. So, there is no need of manual feature extraction. This technique can capable of finding the suspicious score of each program statement. Using this technique, we can collect the large amount of data from the test cases and extract the important features. As pooling layer of CNN model reduces the input size and complexity of the model, so it speeds up the training process. This framework can also be able to calculate the suspicious score of each statement and accordingly assign the rank.