| Summary: | 碩士 === 華梵大學 === 工業工程與經營資訊學系碩士班 === 96 === This research is to improve the previous research model, “Ant-Based Self-Organizing feature Map(ABSOM) Neural Network”, to combine the K-Means into a two-stage clustering method, and further to evaluate the performance of the method using some public databases.
The ABSOM utilizes the pheromone mechanism of ant colony system to memorize the historical process of the best matching units(BMU) selected and adopts the exploitation and exploration state transition rules of the Ant Colony System(ACS). However, the selection of the BMU in the whole algorithm mainly considers the Euclidean distance. In the exploration rule, the capacity of pheromone in each output neuron is simply used to determine the BMU, but not the probability generated from the capacity of pheromone. Additionally, the updating of pheromone for the map neurons ignores the evaporation effect. Thus, this research modifies these two state transition rules for determining the BMU to enhance the ABSOM in order to make the algorithm more complete, and reach the expected performance for cluster analysis.
This research improves the ABSOM algorithm to be an enhanced Ant-Colony Self-Organizing feature Map (eABSOM) and further combine the proposed method with K-Means as a two-stage clustering method. After compared with Kohonen’s Self-Organizing feature Map(SOM) and ABSOM, the eABSOM shows better visualization results(U-matrix) and clustering performance indices.
|
|---|
