A Hybrid Clustering Based on ACO and Single-Pass

摘要

Clustering is an important unsupervised machine learning (ML) method, and single-pass (SP) clustering is a fast and low-cost method used in event detection and topic tracing. To obtain a better result, SP is always needed to run multi-times in previous literatures due to it is dependent on the order of the instances appearance, which is not the global optimum. To tackle this situation, this paper proposed a hybrid clustering based on ant colony optimization (ACO) and SP, which exploits the global optimization ability of ACO and the superiority of SP, and takes the results of SP as a positive feedback implemented in ACO to improve the quality of the clustering. Experiments on two UCI datasets show that it is better than the multi-times running of the SP according to two different evaluation measures, which verifies its effectiveness and availability.