Source code size prediction using use case metrics: an empirical comparison with use case points

Abstract

Software source code size, in terms of source lines of code (SLOC), is an important parameter of many parametric software development effort estimation methods. In this paper, we investigate empirically the early prediction of SLOC for object-oriented software using use case metrics. We used different modeling techniques to build the prediction models. We used the univariate logistic regression and the simple linear regression methods to evaluate the individual effect of each use case metric on SLOC, and the multivariate logistic regression and the multiple linear regression methods to explore the combined effect of the use case metrics on SLOC. We also used in the study different machine learning methods (k-NN, naive Bayes, C4.5, random forest, and multilayer perceptron neural network). The prediction models were evaluated using the receiver operating characteristic analysis, particularly the area under the curve measure, and leave-one-out cross validation. An empirical study, using data collected from five open source Java projects, is reported in the paper. The use case metrics have been compared to the well-known use case points method. Results provide evidence that the use case metrics-based approach gives a more accurate prediction of SLOC than the use case points-based approach.