| Summary: | 碩士 === 國立臺北科技大學 === 生產系統工程與管理研究所 === 90 === The neural networks and genetic algorithms are also feasible for clustering analysis in data mining. Artificial neural networks (ANNs) and genetic algorithms (GAs) have been applied in many areas and obtained very promising results. Kuo and his colleagues have presented that using self-organization feature maps (SOM) network to determine the number of clusters and the starting points and then employing the K-means method to find the final solution, can provide very good solution. Besides, Kuo and his colleagues also proved that K-means can be replaced by genetic algorithms in order to get better result. However, further improvements are still necessary. For instance, in some cases, it is quite difficult to determine the cluster number by observing the outcome of network output array, unless the network topology is very clear. Therefore, this study is dedicated to using adaptive resonance theory II (ART2) network to determine the initial solution and then applying genetic K-means algorithm (GKA) to find the final solution. It is compared with ART2 followed by K-means.
In order to verify the proposed method, data from Monte Carlo Simulation are employed. The simulation results show that the ART2+GKA is really better than the ART2+K-means both in mean within cluster variations and misclassification rate. A real world problem, a recommendation agent system for a web PDA company, is applied. In this system, the browsing paths are used for clustering in order to analyze the customers’ browsing preferences. The results also show that ART2+GKA is much better based on the mean within cluster variations.
|
|---|
