Extending Automatic Design Pattern Detection Tools by Label Correcting Operation

Design patterns (DP) are reliable and stable solutions to frequently occurring problems in object oriented software design. Detecting design patterns can reveal design purposes in code and facilitates maintainance. Thus automatic design pattern detection is an active area or research. Majority of automatic design pattern detection tools have high recall. However they suffer from high false positive rate since some of the design patterns are structureally and/or behaviorally similar. Hence another research direction is filtering the results of other automatic DP detection tools. In this paper, correcting the labels of these false positives is presented for the first time. Several metrics have been proposed that facilitates this process. Automatic label correcting is performed by data mining techniques. Results achieved with this process are as follows: accuracy about 98.2% in "multi-class classifications", 98.2% mode in "one versus all" classification and 100% mode in "pairwise" classification.