Automated selection of a software effort estimation model based on accuracy and uncertainty

Abstract

Software effort estimation plays an important role in the software development process: inaccurate estimation leads to poor utilization of resources and possibly to software project failure. Many software effort estimation techniques have been tried in an effort to develop models that generate optimal estimation accuracy, one of which is machine learning. It is crucial in machine learning to use a model that will maximize accuracy and minimize uncertainty for the purposes of software effort estimation. However, the process of selecting the best algorithm for estimation is complex and expert-dependent. This paper proposes an approach to analyzing datasets, automatically building estimation models with various machine learning techniques, and eval-uating and comparing their results to find the model that produces the most accurate and surest estimates for a specific dataset. The proposed approach to automated model selection combines the Bayesian information criterion, correlation coefficients, and PRED measures.