An improved ACO-based Clustering Algorithm

• Main Authors: Yu-Ching Ting, 丁淯淨
• Other Authors: Yucheng Kao
• Format: Others
• Language: zh-TW
• Published: 2007
• Online Access: http://ndltd.ncl.edu.tw/handle/79400870591214065107

碩士 === 大同大學 === 資訊經營學系(所) === 95 === Data Clustering is one of the important research topics of data mining. Many people used meta-heuristic method to solve this NP-Hard problem. Previously, Shelokar is the first one to propose an ant colony approach for clustering. Because the heuristic information does not be used in this approach, more ants and more CPU time are required to attain the best solution. In this paper we improve this approach by adding heuristic information to the state transition rule. At the core of the algorithm we use both the accumulated pheromone and the heuristic information, the distances between data objects and cluster centers of ants, to guide artificial ants to group data objects into proper clusters. This allows the algorithm to perform the clustering process more effectively and efficiently. Due to the nature of stochastic and population-based search, our algorithm can overcome the drawbacks of traditional clustering methods that easily converge to local optima. Experimental results show that the algorithm is batter than Shelokar’s approach. Finally, we apply our algorithm to solve Cell Formation Problem and have a good result.