| With the constant progress of software industry, software size measurement and software cost estimate have become an improtant part of the software engineering. A variety of estimation methods are also playing an important role in the area of applicable. The principle of statistical analysis provides a scientific data analysis method for different software cost estimation methods, and also lay a solid foundation for predicting software cost. As an important componet of the statistical analysis theory, regression analysis has special advantage that can not compare with the other software estmation methods. Regression analysis has strong objectivity,and can extract valuable information from software historical data comletely. At the same time, it has a very good repeatability.In addition, it has a very sophisticated theoretical system and supported operating evironment. Therefore, regression analysis has been one of commonly used method for the software researchers and estimators.In this paper, the common regression methods of statistical analysis is studied, and the applicability of different regression methods in software cost estimation activity is analysed. Then, a general process of software estimation based on statistical regression methods is presented, and different regression methods in this process is expounded.In this paper, Cook distance, Mahal distance and other statistics are used to pretreat the software history data sets and detect outlier data. Stepwise regression and variance analysis are used to build estimation models, fourtheremore, ridge regression and weighted regression are used to calibrate estimation models. By these methods, the prediction ability of the software estimation models can be improved.Finally, we analysis two different types of software history data sets using regression methods. The experimental results show that the regression modeling method can be suitable for different data sets, and get better prediction results. |
PDF Full Text Request