A Software Defect Prediction Model Based On BP Neural Network Using Improved ACO Optimization

Software defects are inherent properties of the computer software and the main reason that leads to the exception of software system during operation.Therefore,finding the software defects as soon as possible can avoid or minimize the waste of resources caused by fixing software defects during the later stage of software development.Since 1970 s,software defect prediction technology has been developed well and has received great attention in the field of software engineering as the way to predict software defect distribution.A lot of software defect prediction models have been proposed in recent 40 years and achieved good performances as well.Software defect prediction model based on BP neural network is one of the best models.In this modle,various attributes and history data of software defect are used to predict if defects exist in the software modules.Although the software defect prediction model based on BP neural network has a good prediction performance,the model also has some shortcomings in the information of input layer and the optimization of model parameters,which leads to a poor prediction accuracy.In order to improve the prediction accuracy of BP neural network software prediction model,the main contents focused by this thesis are as follows: Firstly,the principal component analysis(PCA)method integrated with the mutual information and self-information was proposed to reduce the dimensions of data set.At the same time,the number of model input elements is minimized under the premise that the original input information is intact.Secondly,the ant colony optimization algorithm based on difference initialization pheromone and local path optimization was proposed to replace the BP neural network's own "gradient descent" method to optimize parameters,which can improve the probability of getting the better parameters.Thirdly,a software defect prediction model based on improved ACO optimization BP neural network was established,which can improve the accuracy of defect prediction.Finally,a ten-fold cross simulation experiment on data set from NASA was conducted to proof the good performance of the proposed model.The experimental results showed that the proposed method could achieve faster convergence rate and better accuracy compared with conventional methods.