Evaluating Thresholds for Object-Oriented Software Metrics

Abstract

Software metrics measure quantifiable or countable software characteristics. Researchers may apply them to provide better product understanding, evaluate the process effectiveness, and improve the software quality. A threshold is a value that aids the proper interpretation of software measurements; it indicates whether or not a given value represents a quality risk. Thresholds are unknown for most software metrics, inhibiting their use in a software quality assessment process. In a previous paper, we proposed a catalog with 18 object-oriented software metrics thresholds, providing a preliminary case study in proprietary software to validate them. This article evaluates these thresholds more deeply, considering significant aspects. We show a new example of threshold derivation, discussing it qualitatively. We explain these software metrics and discuss their threshold values, presenting each one’s application level, definition, formula, and implications for the software design. We conduct a study with two software systems to evaluate the capacity of our thresholds to identify software quality enhancement after a restructuring process. We assess these thresholds using two case studies, comparing the evaluation provided by the thresholds with the qualitative analysis given by manual inspections. The study results indicate that the thresholds may lead to few false-positive and false-negative occurrences, i.e., the thresholds provide a proper quantitative assessment of software quality. This study contributes with empirical evidence that the metrics’ thresholds proposed in our previous work provide a proper interpretation of software metrics and, hence, may aid the application of software metrics in practice.