Comparative analysis of the effectiveness of data mining classification methods: A case of travel agency

In tourism sector, for both tourists and businesses, a data-dependent process is experi-enced from the decision-making stage to the end of a travel. Businesses do all their transactions by using information technologies from the beginning to the end. In the same way, tourists use information technologies throughout the process, from searching for holiday to sharing of experiences after holidays. As a result, large amounts of data are accumulated on the databases of the businesses, and websites and social media related to tourism. The abundance of data, coupled with the need for powerful data analysis tools, causes some problems in reaching the searched information. It leads to inability to reach what a tourist is looking for and the situation of making knowledge-independent decisions in the abundance of data for an enterprise. In order to obtain competitive advantage, it is not important to have the data, but to discovery useful information necessary for the business from the data. In this regard, data mining techniques deal with this problem. Studies have proven that data mining techniques are reliable in discovering useful information from huge amount of data. The aim of this doctoral dissertation is to profile customers who buy and don't buy daily tours using classification techniques from data mining methods. For this purpose, the data -in a database where the transactional data regarding the booking and daily tour are recorded- of A class travel agency operating in Antalya, were used. A booked tourist's daily tour purchase status is "BUYER" or "NON-BUYER", which is a classification problem. Therefore, methods related to the classification function of data mining were used in the study. During the research process, the data in different tables in the travel agency database about reservation, daily tour ticket, products and customers were aggregated in a table. 11 different classification algorithms were run on the preprocessed data. C4.5 algorithm from decision tree classification techniques, which gived the best performance and correct estimation rate among these algorithms, was used for the discovery of customer profiles. Although there are not enough features related to the customer in the travel agency database, in the analysis, some profiles related to the customers who bought the daily tour were reached. It was determined that groups buying the tours according to the region where the daily tour was sold differed, and age group and tourist type (family, group, etc.) were the most related feature for buying daily tours. The research process and findings are expected to show the way to researchers and practitioners working in the relevant field.