Performance evaluation of a code complexity measurement tool: An empirical approach

Abstract

Quality is an important goal desired by stakeholders in every Program Code (PC) developed to meet its functional requirements. However, empirical evidence on tools that measure code complexity and predict future maintenance before deployment is a problem. This work evaluates a Complexity Measurement Tool (CMT) and compares its performance with Code Metrics (CM), an equivalent existing complexity tool using Cyclomatic Complexity Value (CCV), Cyclomatic Complexity Category (CCC) and System’s Computational Time (SCT). The CMT and Code Metric (CM) were subjected to evaluations by conducting two sets of controlled experiments involving each of 25 Student Programmers (SP) and Professional Programmers (PP). The quality of code artifacts obtained from the experiments was assessed using computed complexity values obtained with CMT and CM. Some other metrics obtained from the experiments were Effort/time range for Coding (EC) and Number of Bugs (NB). Data analysis was carried out using descriptive statistics and correlation, tested at  0.05. The mean CCV reported by CMT for SP and PP respectively, were 7.5±0.1 and 9.2±0.1, while for CM were 8.5±0.1 and 9.0±0.1 respectively. The CCC reported by CMT for both SP and PP were ‘risk-free’ and ‘moderate-risk’. The average SCT obtained from CMT and CM for all the codes were 1.0±0.01 and 3.0±0.01 minutes respectively. A strong correlation exists between Line of Code (LoC) and CCV while a weak correlation between NB and CCV as measured by CMT for SP and PP. No CCC was reported by CM; also, no CCV was computed by CM for codes with bugs. The included CCC makes CMT result interpretation better. The speed also gave it an upper hand over the code metrics embedded in Visual Studio environment.