Forecasting Methods, The Svm-based Web Application Defects

Number of faults is a key indicator of software quality. It's also a hotspot of software engineering research. How to decrease number of faults is a problem to any software development organization. Testing is a common way to reduce software faults. But testing is costive and only available when executable product is available. It's too late for development organization to remove all faults when testing is completed.To improve software product quality more effectively under the constraint of limited resources, software fault prediction was created in software engineering research. Hypothesis of software fault prediction is that complex software modules are more likely to have faults than simple one. Statistical metrics can represent the complexity of software modules. So it is possible to predict faults in software modules or whether a module is fault-prone based on statistical metrics that are easy to get. With software fault prediction, software development organization can focus on fault-prone modules. This will help the organization to improve software quality more effectively with limited resources. Now software faults prediction is a proven technology to improve software quality and reliability.This paper classified current research about software fault prediction. Based on this work, this paper proposed to employ Support Vector Machine in software fault prediction. SVM is a kind of forward neural network built with support vector algorithm. As product of statistical learning theory, it has excellent ability of learning and generalization. Currently used metrics can not effectively represent complexity of web applications which is special for its distributed logics. To solve this problem, web application based on Struts web application framework from Apache, which is open source organization, is selected to be object of this research. New metrics suitable for it are created to represent its complexity better.Based on above work, experiments were performed using data from the project CMM Software Quality Assurance Platform sponsored by national 863 plan. Before-mentioned new metrics group and software fault prediction based on SVM are verified. Results of these experiments proved that software fault prediction based on SVM is an effective way to predict faults of web application based on Struts framework.