Software Cost Estimation For E-government Affair At An Early Stage Of Development

For exposing the obvious problems of subjectivity and inconsistency that interference the cost estimation's accuracy in e-Government affair,effectively controlling the government special investment,and creating a relatively fair environment for outsourcing and bidding.It's urgent to establish a scientific evaluation system on investment forecasting.Most of the traditional research focuses on the construction of software cost estimation models at the late stage,The public data set used include the information of the completed software projects,but the whole process is limited by the application scenarios,software types,and so on.Not yet have Appropriate and authoritative software cost estimation models for this.In addition,there are few relevant studies on the cost estimation of software development's early stage due to the difficulties in obtaining and mapping cost information.The research boundary of this paper is e-government application software.The feasible premise is that the e-government system has good similar attributes and the structured development technology is very mature.The technical difficulties are obtaining effective cost characteristics from the declaration with text attributes and get software early cost interval prediction.This paper proposes a software cost estimation applied at an early stage of development.Firstly taking advantage of text processing technology that get rough information obtained from declaration,then bringing these keywords into the novel mapping function to gain the vectorized expression of cost features;Next inputting the vectors into trained clustering model,which can tell us where the software belongs based the principle of minimum distance,actually the true need is to get similar cases witch will be filtered by a threshold later,and we can get the initial estimation by using median method;Finally bringing the vectors into trained error model that can combines T-test to predict the cost interval under a certain confidence interval.The innovations of this paper lies in:(1)Propose the mapping function to convert the rough text information into the uniformed cost feature vector;(2)Use the weighted Gaussian kernel fuzzy clustering algorithm to make up the defect of uncompleted and high-dimensional cost feature,future propose an improved harmony search algorithm that combines genetic cross mutation and chaos theories to dynamically adjust the clustering center of the cost clustering model;(3)Propose to train the noise variance model offline with the support vector regression machine SVR aiming to retrieve the results from previous clustering model,which even can get the software cost estimation under the specified confidence interval.Then using 425 declarations to complete the comparative experiments,it is verified that the method in this paper has good feasibility and effectiveness,and compared with the traditional software cost prediction method,this method has good accuracy.