MUREM: A Multiplicative Regression Method for Software Development Effort Estimation

Abstract

In this paper a multiplicative regression method to estimate software development effort is presented. This method, which we call MUREM, is a result of, on the one hand, a set of initial conditions to frame the process of estimating software development effort and, on the other hand, a set of restrictions to be satisfied by the development effort as a function of software size. To evaluate the performance of MUREM, it was compared with three regression models which are considered as important methods for estimating software development effort. In this comparison a battery of hypothesis and standard statistical tests is applied to twelve samples taken from well-known public databases.These databases serve as benchmarks for comparing methods to estimate the software development effort. In the experimentation it was found that MUREM generates more accurate point estimates of the development effort than those achieved by the other methods. MUREM corrects the heteroscedasticity and increases the proportion of samples whose residuals show normality. MUREM thus generates more appropriate confidence and prediction intervals than those obtained by the other methods. An important result is that residuals obtained by the regression model of MUREM satisfy the test for zero mean additive white gaussian noise which is proof that the estimation error of this model is random.