Outliners detection method for software effort estimation models

Abstract

Outliner detection methods are studied as an approach for simulated in-house dataset creation. In-house datasets are understood as an approach for increasing the estimation accuracy of the functional points-based estimation models. The method which was selected as the best option for outliners’ detection is the median absolute deviation. The product delivery rate was used as a parameter for the median absolution deviation method. The estimation accuracy was compared for a public dataset and simulated in-house datasets, using stepwise regression models. Results show that in-house datasets increase estimation accuracy. © 2019, Springer Nature Switzerland AG.