Estimating Interval of the Number of Errors for Embedded Software Development Projects

Previously, we investigated the prediction of total effort and errors for embedded software development projects using an artificial neural network (ANN). In addition, we proposed a method for reducing this margin of error. However, methods using ANNs have reached their improvement limits, since an appropriate value is estimated using what is known as point estimation in statistics. In this paper, we propose a method for predicting the number of errors for embedded software development projects using interval estimation provided by a support vector machine (SVM) and ANN. In our evaluation experiment, we compared the accuracy of the SVM model with that of the ANN model using 10-fold cross-validation. Results of Welch´s t-test show that the SVM model is more accurate.