A Novel Social Interaction Assistive Device for Arab Deaf People

Abstract

Many deaf people worldwide face problems with integrating into society and interacting with people who do not understand sign language. This can lead to isolation and difficulty in expressing feelings. In this research, our primary goal is to help deaf people communicate, express their feelings, and socialize with others. Toward that end, 40 Arabic words that are commonly used in social interactions were used to build a dataset of hand movements used by deaf people to express these words. These movements were recorded using a Leap Motion Controller ( LMC ). The resulting dataset consists of 1,579 instances and 112 features, recorded with the help of five deaf persons. Feature reduction and oversampling techniques were applied to analyze the dataset. Machine learning algorithms were then used to build a model that is able to classify any given hand posture or gesture into one of those 40 words. This work compared the performance of nine classification algorithms: Random Forest, Decision Table, Classification via Regression, K-Nearest Neighbor (KNN), Simple Logistic, Input Mapped Classifier, Random Tree, J48, and Bayes network. Results show that the Random Forest model achieved the highest accuracy with over 90%, outperforming the other eight models. Subsequently, a usability study was conducted by 10 deaf people to test the effectiveness of the proposed assistive device. The results suggest that the proposed device is useful for facilitating social communication with deaf people. It also suggests that the device was preferred, when compared with other relevant devices.