ImpactScale: Change Impact Metric for Maintenance of Large Software Systems

In software maintenance, changing modules which affect many other modules are intractable. We defined a new metric, ImpactScale, which quantifies the scale of the change impact of a module to improve the accuracy of fault prediction. Both product metrics and process metrics are required to predict faults effectively in maintenance. However, process metrics cannot be always collected in practical situations. ImpactScale is designed to improve the accuracy of fault prediction by using only product metrics under situations without process metrics. The change propagation model for ImpactScale is characterized by probabilistic propagation and relation-sensitive propagation. To evaluate ImpactScale, we predicted faults in two large enterprise systems using Poisson regression and the effort-aware models. The results showed that adding ImpactScale to existing product metrics increased the number of detected faults at 10% effort by over 50%. ImpactScale also improved the predicting model using existing product metrics and dependency network measures.