Employing Support Vector Machines with Genetic Algorithm for Software Reliability Prediction

• Main Author: 陳木龍
• Other Authors: 羅榮華
• Language: zh-TW
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/97542182086922252822

碩士 === 佛光大學 === 資訊學系 === 98 === The size and complex of software development increases gradually following by the high technology coming. The system reliability is concerned more and more, which becomes one of important characteristics of the system. Under that environment, several experts provides growing models in order to measure and predict the software reliability, such as Neural networks, Support vector machines (SVMs)、Timing array analysis and so on. The thesis provides a SVMs and Genetic Algorithm (GA) for predicting software reliabilities., moreover, SVMs has been successfully used to solve the problems of non linear regression and Timing arrays. However, SVMs for predicting reliabilities are still seldom seen. We use SVMs for predicting , then through GA reproduction , crossover , and mutation in order to search the arguments of SVMs. In the part of experiment ,the data is collected by 21-weeks testing dates (T1 system ,Bell Laboratories) and 28-weeks history datas(Q.P.H et al supported) . In the test we divided SVMs’ data into 3 training types (form proportional picking data , accumulative data , handing-history data and recency ). Additionally, the window size of data is considered in the simulation. As the result of the experiment , the model provided by our laboratory is feasible and the prediction ability and accuracy of the comparison with others are higher. The software programmers and testers, therefore, are able to gain the whole ideas of reliabilities in the early period of developing.