On the effectiveness of accuracy of automated feature location technique

Abstract

Abstract-Automated feature location techniques have been proposed to extract program elements that are likely to be relevant to a given feature. A more accurate result is expected to enable developers to perform more accurate feature location. However, several experiments assessing traceability recovery have shown that analysts cannot utilize an accurate traceability matrix for their tasks. Because feature location deals with a certain type of traceability links, it is an important question whether the same phenomena are visible in feature location or not. To answer that question, we have conducted a controlled experiment. We have asked 20 subjects to locate features using lists of methods of which the accuracy is controlled artificially. The result differs from the traceability recovery experiments. Subjects given an accurate list would be able to locate a feature more accurately. However, subjects could not locate the complete implementation of features in 83% of tasks. Results show that the accuracy of automated feature location techniques is effective, but it might be insufficient for perfect feature location.

Supporting Proactive Refactoring: An Exploratory Study on Decaying Modules and Their Prediction

“…We conducted an analysis under the assumption that mitigating the problems of high false positives and low true positives can improve the performance of the prediction model. We artificially modified the result of an IR-based impact analysis technique in two steps, which are inspired by the task input generation approach of a feature location study [24]. First, we decrease the number of false positives by randomly removing false positives from the result.…”