Software Project Cost Estimation

Software cost estimation is the most important chain of software project management and is important basement of software cost management, too. In the former software cost management, the cost estimation methods have some problems, such as subjective experience's effect, uncertain estimation and so on. So far, there is no method can solve these software cost estimation problems perfectly.This paper analyses software cost's component and features, discusses the relationship between software cost estimation and software project management, introduces the software cost estimation's basic steps and train of thought. To solve the problems of uncertain factors and difficulty to decide cost in early period of software project, use neural network and fuzzy theory together to estimate software cost. Construct a fuzzy-neural network cost estimation system, which uses cost related factors as inputs and BP network as basic network structure, add fuzzy theory to deal with the inputs, outputs and middle process. Further more, use PCA method to improve this system. 60 projects'data is used as samples to train the network. Compare the estimation result with projects'real results and COCOMO estimation results. The comparison indicates that the fuzzy-neural network system can estimate software cost more accurate and quicker than COCOMO model. At last, imply this estimation model into an actual case, the result shows that this model can get accurate and quick estimation, which can supply important basis for the real project's process management..The training and test result through historical data and new project data prove that the software cost quick estimation model which mix fuzzy-neural network and PCA together and combine different methods advantages, adapt to the features of software cost. This model can greatly improve the software cost estimation accuracy and effect.