Proposing an Effective Artificial Neural Network Architecture to Improve the Precision of Software Cost Estimation Model

Abstract

Software companies have to manage different software projects based on different time, cost, and manpower requirement, which is a very complex task in software project management. Accurate software estimates at the early phase of software development is one of the crucial objectives and a great challenge in software project management, in the last decades. Since software development attributes are vague and uncertain at the early phase of development, software estimates tend to a certain degree of estimation error. A software development cost estimation model incorporates soft computing techniques provides a solution to fit the vagueness and uncertainty of software attributes. In this paper, an adaptive artificial neural network (ANN) architecture for Constructive Cost Model (COCOMO) is proposed in order to produce accurate software estimates. The ANN is utilized to determine the importance of calibration of the software attributes using past project data in order to produce accurate software estimates. Software project data from the COCOMO I and NASA'93 data sets were used in the evaluation of the proposed model. The result shows an improvement in estimation accuracy of 8.36% of the ANN-COCOMO II when compared with the original COCOMO II.