Oracle Clustering: Dynamic Partitioning Based on Random Observations
Reza Zafarani, Ali Akbar Ghorbani · 2008
In this paper, a new dynamic clustering algorithm based on random sampling is proposed. The algorithm addresses well known challenges in clustering such as dynamism, stability, and scaling. The core of the proposed method isbased on the definition of a function, named the Oracle,which can predict whether two random data points belongto the same cluster or not. Furthermore, this algorithm isalso equipped with a novel technique for determination ofthe optimal number of clusters in datasets. These properties add the capabilities of high performance and reducing the effect of scale in datasets to this algorithm. Finally, the algorithm is tuned and evaluated by means of various experiments and in-depth analysis. High accuracy and performance results obtained, demonstrate the competitiveness of our algorithm.