Sign Language Recognition of Autistic Children using Wearable Sensors Signals Features Selection

Children with autism spectrum disorder (ASD) often face difficulties and challenges in social interactions, communication, and eye contact due to neurological disorders. This neurological disorder causes children to have a lack of social interaction, sensory processing issues, repetitive behaviours, trouble adapting to change, and executive functioning problems. This article proposes the autistic children’s gesture recognition framework using wearable sensors feature selection-based machine learning techniques. The system records and analyzes individuals’ gestures, providing valuable data for understanding their behaviour. A dataset of $\mathbf{2 4}$ gestures from 10 children with ASD was collected, and various machine learning models, such as KNN-Manhattan, KNN-Euclidean, neural networks, and random forests, were trained using time- and frequency-domain features extracted from the sensor data. Cross-validation techniques were applied to evaluate the models’ accuracy, with most classifiers achieving around $\mathbf{8 8 \%}$ and the highest accuracy of about $\mathbf{9 1 . 5 \%}$. This approach demonstrates the potential of wearable technology and machine learning in improving the understanding and support of individuals with ASD.