Analisis Unjuk Kerja K-Nearest Neighbour untuk Klasifikasi Citra Aksara Bali Tulis Tangan

Abstract

 A lack of philologists and the vulnerability of palm leaf material have become triggers for the scripting automation or transliteration of Balinese script images on computer-assisted palm leaves. One possibility to solve this problem is to create a transliteration machine. We proposed a machine learning technique using the k-NN algorithm to create a transliteration of Balinese script images. The benefit of the kNN algorithm is simply working by matching the similarity of new data to the nearest test data.   

Instead of focusing on the classification technique, the study approaches also analyze the two previous processes: the first process is an image preparation process consisting of binarization, cutting the blanks, equalizing size, and thinning. The second is a feature extraction process using the character intensity algorithm. Our experiment employed 18 classes representing 18 Balinese characters.

The optimal accuracy using a 3-fold cross-validation method to 1001 image data yields an average of accuracy is 84.746%. Although the image data used is handwritten, however, kNN algorithm performed classification well using an extensive training dataset. For that reason, the kNN algorithm could be potential for Balinese script images transliteration.