Classification of Business Travelers Using SVMs Combined with Kernel Principal Component Analysis

Abstract

Data mining techniques for understanding the behavioral and demographic patterns of tourists have received increasing research interests due to the significant economic contributions of the fast growing tourism industry. However, the complexity, noise and nonlinearity in tourism data bring many challenges for existing data mining techniques such as rough sets and neural networks. This paper makes an attempt to develop a data mining approach to tourist expenditure classification based on support vector machines (SVMs) with kernel principal component analysis. Compared with previous methods, the proposed approach not only makes use of the generalization ability of SVMs, which is usually superior to neural networks and rough sets, but also applies a KPCA-based feature extraction method so that the classification accuracy of business travelers can be improved. Utilizing the primary data collected from an Omnibus survey carried out in Hong Kong in late 2005, experimental results showed that the classification accuracy of the SVM model with KPCA is better than other approaches including the previous rough set method and a GA-based selective neural network ensemble method.

Supported by the School Research Funding of School of Hotel & Tourism Management, The Hong Kong Polytechnic University and partly supported by the National Natural Foundation of China under Grant 60303012.