APPLICATION OF CONVOLUTIONAL NEURAL NETWORK FOR INDONESIAN SIGN LANGUAGE RECOGNITION: A CASE STUDY AND PHILOSOPHY OF SCIENCE REVIEW

Abstract

The low literacy of Indonesian Sign Language (BISINDO) in the general public remains a barrier to communication with the Deaf community, while research on AI-based sign language recognition generally focuses solely on technical achievements. This study aims to develop a BISINDO alphabet recognition system based on Convolutional Neural Network (CNN) and evaluate it through a philosophy of science perspective. The methods used include collecting a BISINDO alphabet hand image dataset, data augmentation, and transfer learning-based CNN training with the MobileNetV2 architecture and a stepwise training scheme, then deployed to Android using TensorFlow Lite. Test results show the system is able to achieve an accuracy of around 93% on controlled test data with stable real-time inference performance. The scientific contribution of this research is not only in the development of applied AI systems, but also in providing a reflective ontological, epistemological, and axiological framework to assess the validity and social implications of BISINDO recognition technology.