Towards the Definition of Domain-Specific Thresholds
Allan Mori, Eduardo Figueiredo, Elder José Reioli Cirilo · Anais do Simpósio Brasileiro de Sistemas de Informação (SBSI) · 2017
Software metrics provide basic means to quantify several quality aspects of information systems. However, the effectiveness of the measurement process is directly dependent on the definition of reliable thresholds. To define appropriate thresholds, we need to consider characteristics of the information systems, such as their size and domain. There are several studies to propose methods to derive thresholds and evaluate them. However, we still lack empirical knowledge about whether and how thresholds vary across different information system domains. To tackle this limitation, this paper investigates specific thresholds in four information system domains: accounting, e-commerce, health, and restaurant. Our study relies on 40 information systems to derive domain-specific thresholds for 9 well-known software metrics. Our results indicate that lower-bound thresholds (e.g., 15% smaller classes) usually do not significantly vary across domains. However, for all analyzed metrics, upper-bound thresholds (e.g., 5% largest classes) are different in some domains. Moreover, our study also suggests that domain-specific thresholds are more appropriated than generic ones. For instance, we observed in our analysis that the more appropriated threshold to select the 5% largest classes is 290 LOC in health systems and 147 LOC in accounting systems.