Real-time detection of indonesian sign language (ISL) gestures based on long short-term memory

eaf people often encounter communication challenges, and sign language serves as a crucial tool for those who cannot speak. In Indonesia, Indonesian Sign Language (ISL) or Sistem Isyarat Bahasa Indonesia (SIBI) is officially recognized by the government and is taught in Special Schools (Sekolah Luar Biasa - SLB). The sign language dictionary comprises 3483 words, facilitating communication and participation in daily life for the deaf community. This research aims to convert ISL gestures within SIBI into understandable text, employing the Long-Short-Term Memory (LSTM) method as the primary approach. The study conducted experiments with two models: Model 1, using a smaller dataset, and Model 2, employing a larger dataset and implementing the k-fold method. The results indicate that Model 2 with k-fold accuracy achieved an accuracy of 98%, while Model 1 reached an accuracy of 85%. Nevertheless, challenges persist in these models, particularly in detecting words with similar gestures, such as’maaf’ (sorry) and 'cinta' (love), which may still be misidentified. Despite these challenges, this research contributes positively to the development of assistive technology for the deaf community, enabling more effective communication through sign language.