Accurate Estimates without Calibration?

Most process models calibrate their internal settings using historical data. Collecting this data is expensive, tedious, and often an incomplete process. Is it possible to make accurate software process estimates without historical data? Suppose much of uncertainty in a model comes from a small subset of the model variables. If so, then after (a) ranking variables by their ability to constrain the output; and (b) applying a small number of the top-ranked variables; then it should be possible to (c) make stable predictions in the constrained space. To test that hypothesis, we combined a simulated annealer (to generate random solutions) with a variable ranker. The results where quite dramatic: in one of the studies in this paper, we found process options that reduced the median and variance of the effort estimates by a factor of 20. In ten case studies, we show that the estimates generated in this manner are usually similar to those produced by standard local calibration. Our conclusion ...

• Type: Conference
• Year: 2008
• Where: ISPW
• Authors: Tim Menzies, Oussama El-Rawas, Barry W. Boehm, Raymond J. Madachy, Jairus Hihn, Daniel Baker, Karen T. Lum
• Keywords: Accurate Software Process, Historical Data, ISPW 2008, Most Process Models, Software Engineering