Application of function points and data mining techniques for software estimation - a combined approach

Abstract

Project estimation is recognized as one of the most challenging processes in software project management on which project success is depend‐ able. Traditional estimation methods based on expert knowledge and analogy tend to be error prone and deliver overoptimistic assessments. Methods derived from function points are good sizing tools but do not reflect organizations’ specific project management culture. Due to those deficiencies in recent years data mining techniques are explored as an alternative estimation method. The aim of this paper is to present a combined approach of functional sizing measurement and three data mining techniques for effort and duration estimation at project early stages: generalized linear models, artificial neural networks and CHAID decision trees. The estimation accuracy of these models is compared in order to determine their potential usefulness for deployment within organizations. Moreover a merged approach of combining algorithms’ results is proposed in order to increase prediction accuracy and overcome possibility of overfitting occurrence.