Bootstrapping K-means for big data analysis

In recent years, “Big data” has become a popular word in industrial field. Distributed data processing middleware such as Hadoop makes companies to be able to extract useful information from their big data. However, information retrieval from newly available big data is difficult even with the aid of distributed data processing because the task needs many cycles of hypothesis establishment and test due to lack of prior knowledge about the data. K-means algorithm is one of popular algorithms which can be used in earlier stages of data mining because of the algorithm´s speed and unsupervised characteristics. However, with big data, even k-means algorithm is not fast enough to get a desired result in an expected time period. In the paper, we propose a fast k-means method based on statistical bootstrapping technique. Our proposed method achieves roughly 100 times speedup and similar accuracy compared to Lloyd algorithm which is the most popular k-means algorithm in industrial field.