Perbandingan Optimizer Adagrad, Adadelta dan Adam dalam Klasifikasi Gambar Menggunakan Deep Learning

Abstract

Image recognition technology has developed rapidly in recent times. There are many methods springing up in its use. One of them is the Convolutional Neural Network (CNN) as used in this research. The method is used to detect image patterns from the shape of the arrangement of the fingers of one hand as a signal from the identification of the numbers 0 to 9 in SIBI (Indonesian Sign Language System). The problem of the research is that many optimizers emerge in a deep learning method. Therefore, selecting the right optimizer itself is a challenge that can be used as the next reference for input images that do not go through the previous pre-processing stage. The aim of the research is to get the best accuracy score from the comparison of 3 optimizers and their relations to processing time. The conclusion obtained shows that AdaDelta optimizer that has existed for a long time can provide better results than Adam which is the development of the last optimizer.