Bootstrap Prediction Intervals for a Semi-parametric Software Cost Estimation Model
Nikolaos Mittas, Lefteris Angelis · 2009
The vital task of accurate Software Cost Estimation predictions remains a challenging problem attracting the interest of researchers and practitioners. Although Least Squares (LS) regression and Estimation by Analogy (EbA) are two of the most widely applied methods,there seems to be a discrepancy in choosing the best prediction technique. In this paper, we further extend our previous work on the utilization of a semi-parametric model,called LSEbA that achieves to combine the above-mentioned methods. More precisely, we present a method of constructing prediction intervals by the bootstrap resampling technique. The prediction intervals obtained for LSEbA are compared with those of LS and EbA separately,with the aid of a new methodology that takes into account not only the ability of comparative intervals to capture the actual cost, but also their similarity and their width.