| Summary: | 碩士 === 國立高雄第一科技大學 === 資訊管理所 === 98 === This study proposed a group-based particle swarm optimization abbreviated as GPSO, which the swarm is divided into subgroups and each subgroup communicated with each other. The GPSO was applied to the data clustering.
Five types of communication strategies are tested in this study as follows: (1) each swarm subgroup exchange their best particles, (2) each swarm subgroup exchange their 20% of particles which are randomly selected, (3) assigning ranked particles (based on particle’s fitness) to each subgroup, (4) each swarm subgroup exchange their best 20% of particles, (5) combining approaches (2) and (3).
This study used XB indicator, which measures the distance within and between each clusters, as the clustering performance index. This study conducted some experiments using some simulated and public UCI datasets. Our GPSO are better than the traditional PSO and K-mean clustering algorithm. Also, we found the communication strategy of combining the strategy of exchanging the best 20% particles with strategy of assigning the ranked particle performed better than the other four communication strategies.
|
|---|
