Using correlation and accuracy for identifying good estimators
Gary D. Boetticher, Nazim Lokhandwala · 2008
Human-based estimation remains the predominant methodology of choice [1]. Understanding the human estimator is critical for improving the effort estimation process. Every human estimator draws upon their background in terms of domain knowledge, technical knowledge, experience, and education in formulating an estimate. This research presented at the PROMISE 2007 workshop assessed the goodness of human estimation based only on project accuracy. This research extends the goodness of human estimation to also include component correlation. Thus, a good estimator is accurate and also does a good job of ranking component effort. Using this revised definition of goodness of estimation produces an average classification rate of 93.3 percent over 1000 trials. Furthermore, the resulting decision tree is extremely intuitive.