Improving The Results of Learning Nglegena Javanese Handwriting Using Backpropagation Artificial Neural Network

Abstract

The Nglagena Javanese script is one of the cultural assets of the Indonesian nation that needs to be preserved. Various efforts have been made to preserve this script, one of which is using information technology as a learning medium for the Nglagena Javanese script. Information Technology allows the Javanese script to be introduced interactively to students. To support this need, one of which is the ability of information technology to classify Javanese script. Classification of Javanese script is carried out using the Backpropagation Artificial Neural Network (BANN) method. Twenty primary Javanese characters are classified as classes using the Backpropagation Artificial Neural Network (BANN) method. The stages of this research are initial processing, feature extraction, model training, and model testing. Initial processing is carried out to prepare image data so that it is ready for the feature extraction process. The feature extraction method is the Histogram Chain Code (HCC) to obtain the main characteristics of each data class or character of the Nglegena Javanese script. This study compares three research models by adjusting the ratio between the training image data and the test image so that the model that produces the highest accuracy value is produced. The model training and testing process uses 2000 image data, with the percentage distribution of training image data and test images, namely 20%, 80%, second 50%, 50%, and third 80%, 20%, resulting in different levels of accuracy. The results are to produce successive accuracy of 66%, 72%, and 88%.