Software Defect Prediction Model Based On Deep Learning

With the continuous development of modern software,the complexity of the software continues to increase,which leads to errors in software operation.Therefore,effective defect prediction for software is of great significance.In recent years,deep learning is often used for automatic feature generation,and achieved good results.Therefore,this thesis proposes a software defect prediction model based on deep learning.The specific work is as follows:Traditional software defect prediction methods are generally characterized by the static code measurement of the code.Although this method is simple,it is inevitable to ignore the syntax and semantics of the code.This thesis selects the corresponding node according to the code type in the abstract syntax tree and extracts the corresponding features,which provides convenience for subsequent research and experiments.In the field of software defect prediction,the problem of unbalanced data classification is unavoidable.In response to this problem,this thesis studies the current mainstream classification imbalance problem processing methods,and analyzes the principles,advantages and disadvantages of random oversampling and SMOTE methods.In order to apply the AST vector,this thesis proposes a defect sample generation method based on the generation confrontation network.Finally,experiments and comparisons on multiple data sets have confirmed the advantages of the data classification imbalance processing method proposed in this thesis.This thesis combines the advantages of Goog Le Net and Res Net,uses a new convolution module,and builds a deep convolutional neural network to predict software defects.In addition,this thesis uses a logistic regression classifier combined with static code measurement to form integrated learning with the previously constructed model to improve the accuracy of the model.This thesis designs experiments and compares experiments with the commonly used software defect prediction methods.The results show that the model proposed in this thesis can effectively improve the accuracy of defect prediction.