ESTIMATING THE SIZE OF OPEN-SOURCE PHP-BASED APPS BY NONLINEAR REGRESSION MODELS WITH VARIOUS FACTORS
Sergiy B. Prykhodko, Mykhaylo V. Vorona · Collection of Scientific Publications NUS · 2021
The problem of estimating the software size in the early stage of a software project is important because a software size estimate is used for predicting the software development efforts, including open-source PHP-based apps.The purpose of the work is to increase the prediction accuracy of early software size estimation of open-source PHPbased apps.The object of study is the process of estimating the software size of open-source PHP-based apps.The subject of study is the three-factor nonlinear regression models with various factors to estimate the software size of open-source PHP-based apps.To build the three-factor nonlinear regression models we use the technique based on the multivariate normalizing transformations and prediction intervals.These models are constructed based on the Johnson four-variate normalizing transformation for S B family of the non-Gaussian data set from 44 apps hosted on GitHub.The data set was obtained using the PhpMetrics tool (https://phpmetrics.org/).The three-factor nonlinear regression models are built around the metrics of class diagrams: the number of classes, the average number of methods per class, the sum of average afferent coupling and average efferent coupling per class, DIT (depth of inheritance tree) mean per class.To compare the prediction accuracy of the three-factor nonlinear regression models we used the well-known prediction accuracy metrics such as a multiple coefficient of determination R 2 , a mean magnitude of relative error MMRE, and prediction percentage at the level of magnitude of relative error of 0.25, PRED(0.25).The nonlinear regression model constructed around the number of classes, the average number of methods per class, DIT mean per class has the larger PRED(0.25)value and about the same values of R 2 and MMRE that the model in which the third factor is the sum of average afferent coupling and average efferent coupling per class.The scientific novelty of obtained results is that the three-factor nonlinear regression model for estimating the software size of open-source PHP-based apps has been improved by introducing a new factorthe DIT mean per class.This allowed us to increase the PRED(0.25)value by 8%.The practical importance of obtained results is that the software realizing the constructed model is developed in the sci-language for Scilab.Key words: software size estimation; PHP-