PREDICTING BEHAVIOUR CHANGE IN STUDENTS WITH SPECIAL EDUCATION NEEDS USING MULTIMODEL LEARNING ANALYTICS

The presence of educational data in new forms and modes presents new possibilities to students with specialeducation needs (SEN), whose learning and behavior are extremely sensitive to their body states and learningenvironments. Multimodal learning analytics (MMLA) records learner and learning environment data in differentmodalities and processes them to reveal the underlying educational insights. In this paper, we used MMLA inorder to forecast the behavior change of SEN students after enrolling in applied behavior analysis (ABA) therapies,in which ABA therapy is a special education intervention that seeks to treat behavioral disorders and developpositive alterations in behavior. Here we demonstrate that by feeding multimodal educational data, our deep neuralnetwork and machine learning models are able to predict SEN students' behavior change with optimal performanceof 98% accuracy and 97% precision. We also illustrate how environmental, psychological, and motion sensor datacan greatly enhance the statistical performance of predictive models with only conventional educational data. Ourresearch has been utilized in the Integrated Intelligent Intervention Learning (3I Learning) System, augmentingintensive ABA therapies for more than 500 SEN students in Hong Kong and Singapore since 2020.