An In-Depth Comparative Study of Machine Learning Algorithms for American Sign Language Recognition

Sign language recognition represents a significant field of research, with the primary goal of bridging the communication gap between the hearing and deaf communities. The deaf community mostly uses sign language for communication and educational purposes. American Sign Language (ASL) is an important means of communication for the deaf community, and improving communication between deaf and hearing individuals can have a significant impact on their social and professional lives. A promising solution to bridge the gap between these two communities is to use machine learning algorithms to recognize ASL gestures and convert them into alphabet letters. Several machine learning algorithms can be employed for the recognition of ASL, including K-nearest neighbors (KNN), Convolutional Neural Network (CNN), and Support Vector Machines (SVM). This study utilized a comprehensive ASL Image Dataset obtained from MNIST Kaggle. The findings of the comparative study indicate that machine learning algorithms, specifically the CNN model, can enhance communication accessibility for the deaf community. Precise identification and conversion of ASL gestures into alphabetical letters facilitate inclusive and significant exchanges between individuals who are deaf and those who can hear. The results show the CNN model outperformed KNN and SVM algorithms in a study, achieving an accuracy rate of 94%, recall of 93%, precision of 93%, and F1-score of 93%. This is due to CNN’s hierarchical structure, which allows it to identify significant characteristics at various abstraction levels, making it suitable for recognizing intricate and subtle characteristics of ASL gestures.