Building cost estimation models is often considered a search problem in which the solver should return an optimal solution satisfying an objective function. This solution also needs to meet certain constraints. For example, a solution for the estimates coefficients of COCOMO models must be non-negative. In this research, we introduce a constrained regression technique that uses objective functions and constraints to estimate the coefficients of the COCOMO models. To access the performance of the proposed technique, we run a cross-validation procedure and compare the prediction accuracy from different approaches such as least squares, stepwise, Lasso, and Ridge regression. Our result suggests that the regression model that minimizes the sum of relative errors and imposes non-negative coefficients is a favorable technique for calibrating the COCOMO model parameters.

CASE (computer aided software engineering) tools are believed to have played a critical role in improving software productivity and quality by assisting tasks in software development processes since the 1970s. Several parametric software cost models adopt "use of software tools" as one of the environmental factors that affects software development productivity. Several software cost models assess the productivity impacts of CASE tools based only on breadth of tool coverage without considering other productivity dimensions such as degree of integration, tool maturity, and user support. This paper provides an extended set of tool rating scales based on the completeness of tool coverage, the degree of tool integration, and tool maturity/user support. Those scales are used to refine the way in which CASE tools are effectively evaluated within COCOMO (constructive cost model) II. In order to find the best fit of weighting values for the extended set of tool rating scales in the extended research model, a Bayesian approach is adopted to combine two sources of (expert-judged and data-determined) information to increase prediction accuracy. The extended model using the three TOOL rating scales is validated by using the cross-validation methodologies, data splitting, and bootstrapping. This approach can be used to disaggregate other parameters that have significant impacts on software development productivity and to calibrate the best-fit weight values based on data-determined and expert-judged distributions. It results in an increase in the prediction accuracy in software parametric cost estimation models and an improvement in insights on software productivity investments.

Many parametric software estimation models have evolved in the last two decades (L.H. Putnam and W. Myers, 1992; C. Jones, 1997; R.M. Park et al., 1992). Almost all of these parametric models have been empirically calibrated to actual data from completed software projects. The most commonly used technique for empirical calibration has been the popular classical multiple regression approach. As discussed in the paper, the multiple regression approach imposes a few assumptions frequently violated by software engineering datasets. The paper illustrates the problems faced by the multiple regression approach during the calibration of one of the popular software engineering cost models, COCOMO II. It describes the use of a pragmatic 10 percent weighted average approach that was used for the first publicly available calibrated version (S. Chulani et al., 1998). It then moves on to show how a more sophisticated Bayesian approach can be used to alleviate some of the problems faced by multiple regression. It compares and contrasts the two empirical approaches, and concludes that the Bayesian approach was better and more robust than the multiple regression approach.