Software quality evaluation and forecast based on unascertained-SVR model

Abstract

The software quality is a comprehensive fuzzy evaluation problem which might be divided into different layers based on its attributes. An unascertained-SVR measure for evaluating and forecasting the quality of software is established on the basis of analysing the factors affecting the quality risks of software system by applying the unascertained measure theory. Therefore, a new system is found to evaluate software quality based on the knowledge of software project. Then, the quality of software is evaluated and predicted by introducing a new mathematical model - Support Vector Regression SVR model. SVR is one of the best events on dealing with small samples, avoiding the defects of neural network that is easy to fall into local minimum and lower accuracy rate. Finally, the practical application shows that the method overcomes the defect that the variable set by experts, then makes the evaluation results objective and scientific.