Increased Accuracy on Image Classification of Game Rock Paper Scissors using CNN

Abstract

Pandemic COVID-19 made people unable to meet face-to-face and could only do traditional games virtually. One of the traditional games that have shifted into virtual games is the game of rock, scissors, and paper. In order to play virtually, a precise and accurate process of detecting the player's hand gestures is needed. This study aims to facilitate an image classification model with a higher level of accuracy than previous studies to distinguish hand gestures in the form of stone, paper, and scissors. Therefore, in this classification process, one of the methods of Deep Learning is used, namely Convolutional Neural Network (CNN). This research has a contribution in the form of adding epoch values and applying more in-depth hyperparameters to the CNN model. With more epochs and more in-depth hyperparameters, higher accuracy is obtained in detecting hand gestures in the form of stones, scissors, and paper. The dataset used in this study has the title "Rock-Paper-Scissors Images" totaling 2,188 which is divided into 3 classes, namely rock, paper, and scissors classes. This study increased the value of accuracy in previous studies with an increase in the average accuracy of previous studies from 97.66% to 99%.