Enhancing teacher decision making for children with autism using machine learning

Different children with autism have different ways in which they like to be communicated with in relation to different situations – some might prefer gestures, some might prefer words, some might prefer visual images. 

This project is about collecting real time data, using video observation, on interactions where teachers use different types of communication.

This data will be used to develop a Machine Learning Function (MLF) which will help teachers to predict which children are more likely to respond to which type of communication and in which context. So for example, some children might respond better to gestures as opposed to words as a form of communication when being asked to get ready for lunch. 

We hope to develop an MLF which will be able to match child’s characteristics (such as gender and language ability) with the context (such as getting ready for lunch) to predict the best strategy to choose.

This could support teachers to

a) predict what strategy gives the best outcomes for a particular student, or

b) for a new student with a particular profile, predict what strategy is likely to work best for them in achieving a particular outcome.

Potentially, such an approach could help teachers develop effective inclusive practice for children with autism.

We have already tried out the approach with some children in one special school. The project aims to collect further data on more children in more schools (approximately 1,000 observations for each of 20 children), in order to further refine and develop the machine learning function.

This project is currently paused due to the current crisis with novel coronavirus. We hope to continue with data collection, if possible, at an appropriate time in the 2020- 2021 academic year. 

The project is a collaborative effort between IOE and UCL Computer Science.

• Dr Joseph Mintz

• Professor Ingemar Cox
• Dr Vasileios Lampos
• Dr Amanda McCrory
• Dr Carol Rivas
• Dr Xiao Qu