The impact of features on feature location

Presence of large number of Feature Location Techniques (FLTs) poses difficulties when selecting an appropriate FLT, given a software maintenance task. This problem is aggravated by extensive heterogeneity in empirical designs employed to evaluate the FLTs and one such heterogeneity that may feed into conflicting findings across studies, is the feature set sought in evaluations. An analysis of the empirical findings of FL studies suggests that (sought) feature characteristics can have a stronger impact on FLTs performance than differing FLTs. Towards understanding their impact, this paper proposes two feature metrics that are hypothesized as affecting FLTs performances. To evaluate the presented metrics, a controlled experiment on 461 features gathered from four software systems was performed. The focus was to establish the relationship between the metrics and FLT performance. Results of the empirical evaluation suggest that the presented feature metrics strongly impact performance of different FLTs, as measured using established evaluation measures. Thus, this paper facilitates a more standard, transparent selection of feature benchmarks towards fair comparison of FLTs.