A Cost-Effectiveness Metric for Association Rule Mining in Software Defect Prediction

This paper introduces a cost-effective metric for association rule mining in software defect prediction, where defective module conditions are represented as association rules. The metric calculates the expected number of defects discovered in modules that satisfy an association rule, considering a limited test effort or test cases. Since conducting full testing on all modules is impractical due to resource limitations, the proposed metric helps identify the most cost-effective modules to test within given constraints. It is based on the extended exponential Software Reliability Growth Model (SRGM), incorporating the module size parameter to account for the increased effort required to detect defects in larger modules. To evaluate the metric's effectiveness, association rules were extracted and prioritized using the proposed metric across datasets from four open source software projects. The assessment employed the widely used LOC-based cumulative-lift chart to measure the cost-effectiveness of defect prediction. The findings demonstrate that the proposed metric effectively prioritizes rules with a higher potential for defect discovery compared to conventional association rule metrics like confidence and odds ratio.