EVALUATION OF CAMPAIGN CATEGORIES ON KITABISA.COM BY NAIVE BAYES CLASSIFIER METHOD

Abstract

Kitabisa.com is a crowdfunding platform in Indonesia. To help donors choose a campaign that suit their preferences, Kitabisa.com categorizes campaigns manually when campaigners create campaign page. However, there are many options of categories offered so that is possible for campaigners choose wrong campaign category. The Naive Bayes Classifier method can be used to classify campaigns, so it generates recommendations for Kitabisa.com simplifies the campaign categories that can minimize campaigners mistake in choosing categories. Naive Bayes Classifier classifies each campaign based on title, short description, and full description. Document Frequency Improved (DFM) as feature selection implemented for improving the classification performance. This study used 7992 campaign data as training data and 888 campaign data as testing data. The testing applied 5 types of threshold value and using k-fold 10 cross-validations. The best results are shown in the model classification using 5 categories with 3.0 threshold level. The result is an average value of accuracy 90,89%, precision 89.24%, and recall 81.31% .