Klasifikasi dokumen bahasa Indonesia menggunakan metode Dynamic Classifier Selection with Local Accuracies

Abstract

Dynamic classifier selection with local accuracies (DCS-LA) uses k-Nearest Neighbor (kNN) to combine multiple classifiers. The purpose of this research is to compare classification performance using single classifiers (Rocchio, Bayes, and Bernoulli) with DCS-LA combining them. This research classified 249 agricultural documents (3 classes) and 750 news documents (5 classes). The single classifiers obtained different accuracies for the agricultural documents and the news documents. The accuracies for agricultural documents are 52%, 44%, and 61.34% for Rocchio, Bayes, and Bernoulli, respectively, while using news documents resulted in better accuracies, i.e., 81.6%, 86.4%, and 85.6% for Rocchio, Bayes, and Bernoulli, respectively. Similarly, the accuracy of DCS-LA is 56% and 86% for agricultural documents and news documents, respectively. This accuracy is higher than the average accuracy of single classifiers because of the local accuracy calculation in DCS-LA. The accuracy for the news documents is always higher than that of the agricultural documents, because the classes in agricultural documents are highly related. The relation can be seen from the percentage of common words shared among the classes. In the agriculture documents, in average of common words is 48% while it is 10% in the news documents.