Unveiling the Power of Machine Learning and Deep Learning in Advancing American Sign Language Recognition

American Sign Language (ASL) serves as a crucial visual-gestural language for the Deaf community, characterized by its intricate syntax expressed through gestures, facial expressions, and body language. Despite its complexity, ASL presents challenges in the realm of deep learning and machine learning due to its dynamic nature, varied signing methods, and the absence of standardized datasets. The scarcity of labeled data further complicates ASL recognition, hindering the capture of subtle nuances such as hand movements and facial expressions. This research study intends to address these challenges by developing models that convert ASL images into text, thereby advancing the field of sign language recognition. In this study, the efficacy of OpenCV's image classification techniques, namely K-Nearest Neighbors, Naive Bayes, Support Vector Machines, and Logistic Regression, is demonstrated in ASL-related tasks. Remarkably, these models achieve accuracies ranging from 75% to an impressive 80%. The second experiment focuses on ASL-specific applications, leveraging the superior performance of Support Vector Machines (SVM). Utilizing a dataset encompassing four alphabets (A-D), nine words, and ten integers, a combined approach of SVM and YOLOv5 achieves an object detection accuracy of 81.16%, surpassing MediaPipe. Furthermore, the study delves into dataset extension strategies, revealing that a noteworthy accuracy of 92.11% is attained when combining SVM with Convolutional Neural Networks (CNN). Additionally, an impressive accuracy of 98.01% is achieved when utilizing SVM in conjunction with Long Short-Term Memory (LSTM). This comprehensive investigation not only contributes valuable insights for the development of effective sign language communication systems but also significantly expands our understanding of deep learning, machine learning, and ASL identification.