Recognizing Stereotyped Behavior in Children with Autism

Abstract

This project works on helping in identifying and recognizing autistic children's stereotyped behaviors, which can help in diagnosing autism on children. The recognition accomplished by building a signal processing model that collects data from a smartwatch equipped with a gyroscope and accelerometer in order to produce a feature vector of 316 features. This feature vector is used to choose a predictive model with the highest accuracy, which is Ridge classifier in this project. The results show that those common stereotype behaviors could be recognized using the Ridge machine learning algorithm with overall average accuracy ranges between 98.7% to 99.5 %. For hand flapping, head banging, and running back and forth, the overall precision ranges between 98% to 100 %, overall recall ranges between 98% to 100 %, overall F1-score ranges between 98% to 100 % and overall macro, weighted and micro averages is 99 %. This Ridge classifier used to implement a real-time application developed on a smartphone (iPhone) to detect the stereotyped behaviors for autistic children who are wearing the smartwatch (Apple watch)