Empirical study on the distribution of object-oriented metrics in software systems

Abstract

We attempt to model the probability distribution of object-oriented software metrics. We employ 5 distribution models to find out the distributions the metrics follow. We use AIC, BIC and RMSE as goodness-of-fit measures. Though the past studies have shown that the software projects frequently follow power law, having a Pareto distribution, we seek to study more number of software systems and distribution models to infer more generalizable results, since they occasionally seem to follow Log-normal or Gamma distribution as well. Apart from these three models we have also considered Weibull distribution and Generalized Pareto Distribution (GPD). In this study, we have made an attempt to answer the hypothesis that the object-oriented software metrics follow a particular distribution by comparing various distributions applied over a large number of projects using a recognized statistical framework.