Improvement of the Firefly-based K-means Clustering Algorithm

摘要

The paper proposes two new hybrid global k-means cluster algorithms, the probabilistic firefly k-means (PFK) and the greedy probabilistic firefly k-means (GPFK), to improve the existing firefly k-means (FK) algorithm. The new algorithms combine the firefly optimization algorithm and k-means clustering algorithm for better performance. To be more specific, the optimization algorithms offer the better initial centroids for the k-means clustering algorithm. To test the performance of proposed PFK and GPFK algorithms, experiments are performed on Apache Spark platform based on Map/Reduce framework. The experimental results demonstrate higher clustering accuracies are achieved over the FK algorithm.