Software Cost Estimation Techniques And Research

Software cost estimate is the most important basement of software development. Software developer can know how much they will spend on a project with this process, on the other hand, if we have no software cost estimate during we develop a project, we may face failure or risks The nature of variety of software development and nonlinearity relationship between the software cost and the driven factors of software cost makes us unable to know exactly the rule between the software cost and driven factors of software estimate. Many tools and techniques as we have in such area, there is almost none to make us satisfy because the poorly ability of them. In this paper, a software cost estimate system named Neural-COCOMO was introduced,this system is to construct a neural network base on COCOMOII early design model. Neural network was introduced in the field of software cost estimate for many years, but the accuracy of it is never made it to be useful in practice. The main reason of it is only because that we have no so much history project data to train the network. Neural-COCOMO software cost estimate system is a neural network based on early design COCOMOII. The network was trained twice, the first time of it is trained by the data sampling on the function of early design COCOMOII, This network will approach the function if the number of sampling points is large enough, then , the network will be trained by history data which was provide by COCOMO develop organization ; At last, the Neural-COCOMO software cost estimate system would be used to estimate the real project, the test result of it show that this new system is more accuracy than the early design COCOMOII.