An Improved CURD Algorithm for Source Code Mining

Source code mining algorithm should have the ability to cope with large volume of data and nominal attributes, which are two major characteristics of source code dataset. K-means algorithm is not suitable for clustering source code as it is generally difficult for the users to determine the count of clusters for a previously unknown dataset. CURD clustering algorithm works efficiently. However, it can´t process nominal attributes. In this paper, we propose NCURD algorithm for clustering source code by making CURD applicable to nominal attributes, and by improving the working efficiency of CURD. The experimental results show that NCURD algorithm has excellent clustering performance for clustering source code.