A parallel k-means clustering initial center selection and dynamic center correction on GPU

K-means clustering algorithm is a partition based clustering algorithm which has been widely used in data mining applications. This algorithm suffers from an issue, named initial centers selection. This problem significantly effects on the quality and running time of clustering. Several literatures discussed on this problem and try to select the best initial centers to prevent final results from getting into local minimum and inaccurate results. Although initial center selection decreases the total running time, it imposes a time overhead that can be solved by parallel design. In addition, previous solutions didn't consider the algorithm behavior after selecting the initial centers, which is considered by dynamic correction in this work. Graphic Processing Unites has several parallel cores which provide a parallel device for developers. This paper proposes a parallel initial centers selection and dynamic center correction on GPU which is fast, accurate and scalable.