Pengenalan Bahasa Isyarat Hijaiyah: Augmentasi Data dengan EfficientnetB7

Tanwir Tanwir (Program Studi Magister Ilmu Komputer, Universitas Bumigora)
Husain Husain (Program Studi Magister Ilmu Komputer, Universitas Bumigora)
Rifqi Hammad (Program Studi Rekayasa Perangkat Lunak, Universitas Bumigora)
Andi Sofyan Anas (Program Studi Ilmu Komputer, Universitas Bumigora)
Muhammad Azwar (Program Studi Ilmu Komputer, Universitas Bumigora)

Jurnal Teknologi Informasi dan Multimedia
Vol. 7 No. 4 (2025): November

Sign language plays an important role as the primary means of communication for individuals with hearing impairments. This study aims to improve the accuracy of hijaiyah sign language detection through the application of the EfficientNetB7 architecture and data augmentation tech-niques. The method used, namely the EfficientNetB7 algorithm, was chosen as the base model be-cause of its ability to balance high accuracy with optimal resource utilization by performing data augmentation with rescale, shear, zoom, rotation, and flip horizontal techniques applied to enrich the variation of the original dataset of 6,811 images to 105,615 images. The experimental results show that the combination of EfficientNetB7 and data augmentation produces 99% accuracy on the test data, with consistent performance seen from the confusion matrix and accuracy loss graph for 50 epochs. This study proves that this approach not only improves model generalization but also reduces the risk of overfitting, thus potentially supporting social inclusion through efficient and reliable technology.