A MATHEMATICAL MODEL FOR CODE METRICS INFORMATION PROCESSING OF DATA SCIENCE AND MACHINE LEARNING JAVA APPLICATIONS TO ESTIMATE THEIR SIZE
Oleksandr S. Oriekhov, Тetyana A. Farionova · Collection of Scientific Publications NUS · 2025
The aim of the study is to increase the accuracy and reliability of code metrics information processing to estimate the size of Data Science (DS) and Machine Learning (ML) JAVA applications in the early stages of software project planning using class diagram metrics by building a nonlinear regression model for further use in parametric models of software development effort estimation.Accurate software size estimation is necessary for project planning, resource allocation, and costing in software development.Methods.The research methods include analysis of existing regression equations and models for estimating the KLOC of JAVA applications; collection and processing of information from the code metrics of JAVA applications in the DS and ML areas from the training and testing datasets; ; methods for iterative nonlinear regression models constructing, including methods of probability theory, mathematical statistics, multivariate statistical analysis, linear and nonlinear regression analysis; etc. Results.A five-factor nonlinear regression model for processing information from code metrics for early KLOC estimation of DS and ML JAVA applications and its prediction interval were constructed.The obtained quality criteria assessments of R 2 , MMRE, and PRED (0,25) for the nonlinear regression models on the basis of the initial training sample and the test sample show a high level of accuracy in estimating the KLOC parameter using the obtained five-factor regression compared to the existing regression models and the constructed four-factor regression.The obtained prediction interval is 19.2% shorter than the prediction interval of the built four-factor regression.Scientific novelty.Firstly, a five-factor nonlinear regression model was constructed on the basis of multivariate non-Gaussian data from DS and ML JAVA applications.Firstly, the metric of the total quantity of classes and interfaces was decomposed into separate metrics of classes and interfaces.The practical significance of the obtained results allows us to recommend the five-factor model for use in practice to process information from code metrics for early size estimation of DS and ML JAVA applications.