Software effort estimation using classical analogy ensembles based on random subspace

Abstract

Software effort estimation is one of the important and complex tasks in software project management. It influences almost all the process of software development such as: bidding, planning, and budgeting. Hence, estimating the software project effort in early stages of the software life cycle is considered the key of success of any project. To this goal, many techniques have been proposed to predict the effort required to develop a software system. Unfortunately, there is no consensus about the single best technique. Recently, Ensemble Effort Estimation has been investigated to estimate software effort and consists on generating the software effort by combining more than one solo estimation technique by means of a combination rule. In this paper, we have developed different homogeneous ensembles based on combination of Random Subspace method and Classical Analogy technique using two linear rules over seven datasets. The results confirm that the Random Space ensembles outperform the solo Classical Analogy regardless of the dataset used and that the median rule generates better estimation than the average one.