Neural Networks for Accurate Estimation of Software Metrics

Abstract

Nowadays, software requirements are widely increasing and modern software technologies are rapidly growing. Therefore, planning and managing the software projects are more important as compared to the past. Unsuitable project planning has been the main reason of software project fails in recent years. Estimating is performed at the early stages of project and it is one of the most important activities in software project planning. During the recent years many different methods have been proposed to estimate the software metrics. Software development effort and software size are most important metrics in this field. Since uncertain nature of software projects makes it difficult to estimate the metrics, soft computing techniques have been widely used in software metrics estimation. These techniques can improve the accuracy of estimations in software projects by means of neural networks, genetic algorithm, fuzzy logic and so on. Among all mentioned methods, due to high flexibility and adaptability, neural networks have been used more than the other methods. A comparative study can be useful to discriminate the performance of neural networks regarding the software metrics estimation. In this paper 15 previous research works in term of using neural networks for software estimation were investigated basically. Analyzing the selected research works showed that neural networks can estimate the software metrics more accurate than most common algorithmic methods.