Performance Analysis of Sign Language Recognition using Deep Neural Network Architecture

Sign Language Recognition (SLR) plays a crucial role in facilitating communication for individuals with hearing and speech impairments. Advances in deep learning have significantly improved the automation of gesture recognition. There are many methods used to improve the performance of SLR. This paper evaluates the performance analysis of five models namely Custom CNN, InceptionV3, DenseNet121, Xception, and NASNetMobile on a benchmark dataset, employing preprocessing techniques and optimized training strategies. Among the models, DenseNet121 and Xception demonstrate the highest accuracy, while NASNetMobile offers a strong balance between performance and computational efficiency. The lightweight Custom CNN also delivers competitive results. Statistical analysis supports these outcomes, highlighting the importance of balancing accuracy, speed, and resource usage for practical deployment. Future directions include exploring transformer-based models, integrating multimodal learning, and optimizing for edge devices to promote more inclusive communication technologies.