Visitor Decision System in Selection of Tourist Sites Based on Hybrid of Chi-Square And K-NN Methods

Abstract

Madura Island is one of the islands with a lot of tourism spread over four districts, such as natural, religious, and cultural tourism. And every year, various visitors visit various tourist sites in Madura, so an increase in the number of visitors has been found in multiple places. This is influenced in addition to the type of tourist attraction but also changes in tourist behavior in making decisions to visit tourist objects. Most of the researchers have applied the right decision-making with intelligence-based measurement. However, the accuracy obtained has not yet reached the optimal solution. Therefore, this study uses the Chi-Square and K-Nearest Neighbors (K-NN) methods to recommend tourist attraction locations based on visitor characteristics to increase visitor attractiveness in tourist attractions scattered in Bangkalan, Madura. Chi-Square is used to select features that affect tourist attraction visitor factors by testing the relationship between the variables involved. Meanwhile, K-NN is a method of classifying potential visitor attractions based on their characteristics by using the closest membership calculation, which is the largest from the test data. The calculation is carried out by the square of the Euclidian distance from each object, then sorted from the smallest to the largest value and looking for the value of k as the result of the decision. There are ten features used in the classification, such as tourism type, management services, facilities, gender, age, occupation, education, visitor status, ticket prices, and sales trends. There are three classes classified: low, medium, and high visitor attractiveness. The contribution of this study is to analyze the effect of the characteristics of tourist attraction visitors on increasing visitor attractiveness using the chi-square and K-NN methods. Based on the results of system testing using K-Fold Cross Validation with five folds from 315 datasets, it produces the highest accuracy at k-fold = 3 worth 84.12% with eight selected features.