Computing optimal cocomo effort multiplier values and optimal casebase subsets using monte carlo methods
Robert Joseph Maidhof · Journal of International Crisis and Risk Communication Research · 1996
There have been many studies performed and techniques applied to solve the problem of estimating man-month effort for software projects. Despite all the effort expended to solving this problem the results achieved from the various techniques have not been embraced by the software community as very reliable or accurate. This thesis uses Monte Carlo methods to obtain optimal values for COCOMO effort multipliers which minimize the average of the absolute values of the relative errors (AARE) of man-month estimate for two industry supplied casebases. For example, when using three COCOMO cost drivers (complexity, language experience, application experience) and the COCOMO effort multiplier values, AARE values were 60% for casebase 1 and 53% for casebase 2; using Monte Carlo to obtain optimal effort multiplier values, AARE values were 34% for casebase 1 and 41% for casebase 2. By repeatedly removing the cases which contributed the greatest Absolute Relative Error, the Monte Carlo method was also used to determine optimal casebase subsets with AARE values of less than 10%. This latter approach identifies casebase cases for which the cost drivers may have been rated incorrectly or cases which are not rated consistently with respect to a subset of cases.