Using a Clustering Genetic Algorithm to Support Customer Segmentation 413 to O.l. This study performs the crossover using a uniform crossover routine. The uniform crossover method is considered better at preserving the schema, and can generate any schema from the two parents, while single-point and two-point cross- over methods may bias the search with the irrelevant position of the features. For the mutation method, this study generates a random number between 0 and l for each of ne features in the organism. If a gene gets a number that is less than or equal to the mutation rate, then that gene is mutated. As the stopping condition, only 5000 trials are permitted 4 Experimental Design and Results 4.1 Research data This study makes use of the public data from the 9th GVU's www Users Surve of which concerns with the Internet users'opinions on on-line shopping [5] Velido, Lisboa and Meehan [ll, we selected 44 items about general opinion of www for shopping and opinion of Web-based vendors. Total observations were 2180 Internet users 4.2 Experimental Design We adopt three clustering algorithms- simple K-means, SOM and Ga K-means-to our data. We try to segment the Internet users into 5 clusters. In the case of SoM, we set the learning rate( a )to 0.5 Simple K-means was conducted by SPSS for Windows 11.0 and SOM by Neu- roshell 2 R4.0. GA K-means was developed by using Microsoft Excel 2002 and Pali- sade Softwares Evolver Version 4.06. And, the K-means algorithm for ga K-mear was implemented in VBA (Visual Basic for Applications)of Microsoft Excel 2002 4.3 Experimental Results As mentioned in Section 2.3, intraclass inertia can be applied to compare the per- formance of the clustering algorithms. The intraclass inertia of the three clustering algorithms is summarized in Table 1. Table 1 shows that the performance of Ga K means is the best among three comparative methods Table 1. Intraclass inertia of each clustering method Clustering method K-means GA K-means Intraclass inertia 108504610847091084370 addition, Chi-square procedure also provides preliminary information that helps to make the Internet users associated with the segments a bit more identifiable 3. In this study, Chi-square analysis was used to demographically profile the individuals associated with each of the segments. Table 2 suggests the results of Chi-square analysis. The result reveals that the five segments by all three clustering methods differed significantly with respect to all of demographic variables. So, considering above results, we conclude that GA K-means may offer viable alternative approachUsing a Clustering Genetic Algorithm to Support Customer Segmentation 413 to 0.1. This study performs the crossover using a uniform crossover routine. The uniform crossover method is considered better at preserving the schema, and can generate any schema from the two parents, while single-point and two-point cross￾over methods may bias the search with the irrelevant position of the features. For the mutation method, this study generates a random number between 0 and 1 for each of the features in the organism. If a gene gets a number that is less than or equal to the mutation rate, then that gene is mutated. As the stopping condition, only 5000 trials are permitted. 4 Experimental Design and Results 4.1 Research Data This study makes use of the public data from the 9th GVUís WWW Users Survey, part of which concerns with the Internet usersí opinions on on-line shopping [5]. Like Velido, Lisboa and Meehan [11], we selected 44 items about general opinion of using WWW for shopping and opinion of Web-based vendors. Total observations were 2180 Internet users. 4.2 Experimental Design We adopt three clustering algorithms ñ simple K-means, SOM and GA K-means ñ to our data. We try to segment the Internet users into 5 clusters. In the case of SOM, we set the learning rate (α ) to 0.5. Simple K-means was conducted by SPSS for Windows 11.0 and SOM by Neu￾roshell 2 R4.0. GA K-means was developed by using Microsoft Excel 2002 and Pali￾sade Softwareís Evolver Version 4.06. And, the K-means algorithm for GA K-means was implemented in VBA (Visual Basic for Applications) of Microsoft Excel 2002. 4.3 Experimental Results As mentioned in Section 2.3, intraclass inertia can be applied to compare the per￾formance of the clustering algorithms. The intraclass inertia of the three clustering algorithms is summarized in Table 1. Table 1 shows that the performance of GA K￾means is the best among three comparative methods. Table 1. Intraclass inertia of each clustering method Clustering method K-means SOM GA K-means Intraclass inertia 10850.46 10847.09 10843.70 In addition, Chi-square procedure also provides preliminary information that helps to make the Internet users associated with the segments a bit more identifiable [3]. In this study, Chi-square analysis was used to demographically profile the individuals associated with each of the segments. Table 2 suggests the results of Chi-square analysis. The result reveals that the five segments by all three clustering methods differed significantly with respect to all of demographic variables. So, considering above results, we conclude that GA K-means may offer viable alternative approach