THE NON-LINEAR REGRESSION MODEL TO ESTIMATE THE DEVELOPMENT DURATION OF JAVA APPLICATIONS FOR THE MIDRANGE PLATFORM

Abstract

Data from the ISBSG company shows that the midrange computer platform occupies a software share of approximately 25%. At the same time, the Java language is mainly used for development on this platform. Java applications for the midrange platform have such characteristics as a large size, a significant set of used components, and special requirements for stability. For this reason, performing a reliable assessment of the duration of the development of such applications is an important task, the solution of which has scientific and practical interest.

Analysis of modern models for the duration estimation of development of software applications was done. The most used models for estimating the duration of software development are COCOMO and ISBSG: nonlinear regression equations for estimating the duration of software development depending on development effort. Only the ISBSG model takes into account the features of the platform for which the software is being created. However, this model does not take into account the programming language used to develop the application.

A non-linear regression model was built for duration estimation of applications development written in Java for the midrange platform depending on effort, by creating non-linear regression equation, bounds of the confidence interval and of the prediction interval. The specified model was built using statistical data of the ISBSG repository from 129 projects and the appropriate method based on normalizing transformations. This method was used because empirical data on the duration and effort of software development have a distribution law that differs from the Normal distribution. The data normalization required by this method is done using a decimal logarithm. In the process of examining the data for outliers, 29 of the 129 applications were removed. Better values for the characteristics of the coefficient of determination, MMRE and the percentage of prediction PRED(0.25) were obtained for created model compared to the ISBSG model. The construction of the specified model made it possible to improve the reliability of the obtained duration estimates of the applications development written in Java for the midrange platform.