| Summary: | 碩士 === 國立臺灣科技大學 === 資訊管理系 === 93 === How to estimate software effort more accurately is always a key to the success of a software project. Under-estimating the effort needed for software development may cause the sacrifice of the software quality and seriously lead to the failure of the software development project because of insufficient distribution of the allocated resources. However, over-estimating the software development effort may also cause the problem of the inefficient usage of allocated resources and further lose the chance of gaining the software project in the price bidding because of allocating too many resources. To date, the researches on the effect of the accuracy of software effort estimates by clustering software project data have only concentrated on a single software effort driver. This paper aims at investigating the effect of accuracy of software effort estimation model that is built by using the homogenous software project data. By clustering the project data with single and multiple software effort drives, we use Neural Network and Regression Analysis methods to construct the individual software effort estimation models, and then compare their accuracies with supervised clustering and unsupervised clustering methods of data mining. The results show that: (1) the accuracies of both single and multi-driver software effort estimation models are better than the unclustered model with unsupervised clustering data, but supervised clustering data does not. Therefore, we consider that the property and distribution of software project data influences the accuracy of estimation result. (2) No matter if the software project data are clustered with supervised learning or unsupervised learning, the accuracy of Neural Network software effort estimation model is better than the Regression Analysis model.
|
|---|
