Software Failure Prediction Based on Game Theory and Convolutional Neural Network Optimized by Cat Hunting Optimization (CHO) Algorithm

Predicting the failure of software projects in the stages of software production reduces the losses of software companies. Deep learning methods are an efficient tool for software failure prediction. Imbalance of data sets, intelligent feature selection, and accurate deep learning techniques are among the challenges of deep learning methods for accurate software failure prediction. This manuscript presents an improved game theory method based on the Generative adversarial (GAN) network to predict software failure and success. In the second stage, the cat-hunting algorithm is used to select features and reduce the dimensions of the samples. The dimensionally reduced samples are converted into RGB color images in the third step. RGB images are used for convolutional neural network(CNN) training. The advantage of the proposed method is to intelligently select the feature, reduce the input of the CNN neural network, and simultaneously balance the training samples with game theory to increase the accuracy of the prediction model. In this manuscript, the NASA dataset is used to predict the failure of software projects. The accuracy of the proposed method (GCV) in predicting the failure of software projects is equal to 96.69%. The GCV method is more accurate in predicting the failure of software projects than the LSTM and VGG16 methods. The proposed method is more accurate in feature selection than Chi, IG, and ReF WOA, HHO, and JSO algorithms methods. Predicting the failure of software projects in the stages of software production reduces the losses of software companies. Deep learning methods are an efficient tool for software failure prediction. Imbalance of data sets, intelligent feature selection, and accurate deep learning techniques are among the challenges of deep learning methods for accurate software failure prediction. This manuscript presents an improved game theory method based on the Generative adversarial (GAN) network to predict software failure and success. In the second stage, the cat-hunting algorithm is used to select features and reduce the dimensions of the samples. The dimensionally reduced samples are converted into RGB color images in the third step. RGB images are used for convolutional neural network(CNN) training. The advantage of the proposed method is to intelligently select the feature, reduce the input of the CNN neural network, and simultaneously balance the training samples with game theory to increase the accuracy of the prediction model. In this manuscript, the NASA dataset is used to predict the failure of software projects. The accuracy of the proposed method (GCV) in predicting the failure of software projects is equal to 96.69%. The GCV method is more accurate in predicting the failure of software projects than the LSTM and VGG16 methods. The proposed method is more accurate in feature selection than Chi, IG, and ReF WOA, HHO, and JSO algorithms methods.

Software Failure Prediction Based on Game Theory and Convolutional Neural Network Optimized by Cat Hunting Optimization (CHO) Algorithm Keywords:

Software failure prediction , Feature selection , Cat Hunting algorithm , GAN , SMOTE , CNN