Assessment and Enhancement of Real-Time Recognition of Sign Language Alphabets Through Diverse Machine Learning Techniques

• Published in: 2025 8th International Conference on Data Science and Machine Learning Applications (CDMA) pp. 85 - 90
• Main Authors: Ali, Mubashir, Hosseini, Seyed Ebrahim, Pervez, Shahbaz
• Format: Conference Proceeding
• Language: English
• Published: IEEE 16.02.2025
• Subjects: AdaBoost; Application; Auditory system; Bayes methods; Computational modeling; Computer Vision; Hands; K-Nearest-Neighbor (KNN); Landmark; Logistic regression; Machine Learning; Machine learning algorithms; Multilayer Perception; Naïve Bayes; PSL; Python; Random forests; Random-Forest; Sign language; Support Vector Machine; Support vector machines; Vectors
• DOI: 10.1109/CDMA61895.2025.00020

Communication is at the core of all human interactions, whether personal or professional. It is one of the basics for survival in a community. Verbal communication cannot occur without a coherent, mutually recognized language. According to estimates, deaf people make up approximately 7.4% of the disability population in Pakistan. The prevailing technique emphasizes oralism (teaching spoken language). As a result, reducing the gap in communication is crucial amongst the general population with those who have difficulty speaking. The goal is to create an application that evaluates a user's skills as they cycle through each letter of the Urdu alphabet in Pakistan Sign Language (PSL). Furthermore, this work investigates various ML techniques for recognizing hand motions that involve the detection of landmarks. For the vision of the computers, recognizing locations is customary on certain locations on an item for extracting related attributes from it, such as various people's hands. Machine learning techniques are evaluated, for example, Naïve Bayes, K-Nearest Neighbor, AdaBoost, Support Vector Machine, Decision Trees, Logistic Regression, and Random Forest Classifier. All models were trained and tested on a dataset of 10,000 computed hand data expressions. The model was trained using the dataset in CSV format. The project's goal is to make PSL more accessible and improve communication between the deaf and hearing communities. Multiple trials will be carried out, comprising changes to the models' architectural parameters to attain the maximum possible identification accuracy.