Deaf individuals in Indonesia face significant challenges in communicating with the general public due to limited access to Indonesian Sign Language (BISINDO). The public’s lack of understanding of sign language contributes to stigma and widens social gaps. Therefore, a system capable of automatically and accurately detecting BISINDO is needed. This study aims to develop a BISINDO detection system using the Convolutional Neural Network (CNN) algorithm and the Convolutional Block Attention Module (CBAM). The dataset used consists of 9.169 BISINDO alphabet images from Kaggle, which were augmented to 600 images per class. The system development follows the CRISP-DM stages, from problem understanding to model evaluation. Four training scenarios were tested with varying parameters such as learning rate, epochs, and the use of CNN and CBAM. The best result was obtained from the scenario combining CNN and CBAM (learning rate of 0.001 and 50 epochs), achieving 94% accuracy on training data, 88% on validation, and 87% on testing. The addition of CBAM was proven to improve accuracy by 5-11%. The resulting model demonstrates stable performance and holds potential to be implemented as a communication aid for individuals with hearing disabilities.Kata kunci: Bisindo, Convolutional Neural Network, Convolutional Block Module Attention, CRISP-DM
Keywords: Indonesian Sign Language, Convolutional Neural Network, Convolutional Block Module Attention, CRISP-DM
x