A Study of Software Project Development Cost Control with Genetic Algorithm and General Regress Neural Network

• Main Authors: Chia-Hsiang Chang, 張家翔
• Other Authors: Chang-chun Tsai
• Format: Others
• Language: zh-TW
• Published: 2008
• Online Access: http://ndltd.ncl.edu.tw/handle/50322046545542631796

碩士 === 國立成功大學 === 資訊管理研究所 === 96 === Information Systems can increase company’s competitive strength in the information age. Therefore, information software will keep planning and developing. However, if the software development process without appropriate management, that will prone to extend development time, low quality, improper human resources assignment. This will affect software management and maintenance procedures. So, accurate software cost estimates is the base of budget, time, human resources, and equipment allocation. Traditionally, Lines of Code (LOC) and Function Points (FP) are usually applied to estimate software cost. But these two methods have been unable to obtain good results. During recent years, many studies use Back Propagation Network to estimate software cost and all the studies are better than the traditional method of the accuracy of software cost prediction. Compared with the General Regression Neural Network (GRNN) and Back Propagation Network (BPN), GRNN have advantages included learning speed fast, stable results, and less parameters. The purpose of this study is to use Genetic Algorithms (GA) integrated GRNN. This research tries to find the most suitable smoothing parameter values of the GRNN, and enhance the ability of forecasting. By Comparison with software development costs estimated method in related literature, the accuracy of forecasting model in this research is better than Classification and Regression Trees (CART), Artificial Neural Network (ANN), Ordinary Least Squares Regression (OLS), Analogy-Based Model, and Adjusted Analogy-Based Model.