Filling the gaps: Imputation of missing metrics’ values in a software quality model

Abstract

Hierarchical software quality models usually rely on a number of metrics, which, once aggregated, provide an overview of selected perspectives of a system’s quality. Missing values of some metrics, that usually result from data unavailability, can seriously affect the final score. In the paper we empirically validate a few imputation methods in context of a custom Géant-QM framework, used for evaluation of several open source systems. Early results indicate imputing a missing value based on its close neighbors as data donors introduces less noise that using a wider set of donors.