Software effort estimation using machine learning methods

Abstract

In software engineering, the main aim is to develop projects that produce the desired results within limited schedule and budget. There are many factors that affect the budget of a project. The most important factor that is affecting the budget of a project is effort which includes the developers, managers, and architects working on a project. Estimating effort is crucial since hiring more people than actually needed leads to loss of income and likewise hiring less people than actually needed leads to delay in product delivery. To balance schedule and budget, the effort needs to be correctly predetermined. In software engineering this problem is called as effort estimation problem. Software companies usually use expert judgment to estimate the required effort, however, they are far from getting satisfactory results. The main objective of this research is making an analysis of software effort estimation to overcome problems related to budget and schedule overruns. Our proposed solution not only brings another point of view into software engineering cost and effort estimation but it also tries to improve the software effort estimation process. If practitioners have more accurate estimations then they would be able to manage risks and prevent any losses that may have occurred due to these risks. We have proposed a model that uses machine learning methods such as Multilayer Peceptrons, Radial Basis Functions, Decision Trees, Support Vector Machines and Principal Components Analysis for a better cost and effort estimation in software development projects. We have obtained the metric data that is used as an input to these methods from third parties such as National Aeronautics and Space Administration (NASA) and University of South California (USC) and various projects from software development organizations in Turkey.