SBIR Phase I: Using Data Mining to Optimally Customize Therapy for Individuals with Autism

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) Phase I project includes innovations in data mining and the treatment of autism. Applied Behavior Analysis (ABA) therapy is the gold standard in treating autism. Applying data analytics to data from ABA therapy sessions will contribute in several important ways: a) patterns may be discerned across individuals with autism to better understand variations in autism and create therapies to target these differences; b) patterns may be matched with other data, such as genomic data, to identify cross-patterns that may be useful in better understanding autism and ways to improve therapy; and c) the frontiers of data mining will be expanded to provide guidance in real time. This project will have the following societal impacts: 1) many more individuals with autism across the globe will receive early, quality, cost-effective treatment regimens that will enable them to live more fulfilled lives and reach their full potential; 2) families whose children are good candidates for treatment and receive it will experience reduced stress and better family life; and 3) the additional lifetime cost of not effectively treating children with autism, which is approximately ten-fold the cost of treatment, will be reduced. The proposed project is to extract informative sequential patterns from trial sequences of an individual student, use them to accurately predict trial outcomes, and utilize the predictive model to provide individualized recommendations about how to modify trials and steps of student training. To achieve this goal, predictive data mining will be used. To develop accurate predictive models, the project will build on a large body of recent work in machine learning on temporal predictive modeling and sequential pattern mining, including some of the previous results of the project team. Special attention will be paid to the recent work in educational data mining and intelligent tutoring. Specific key objectives include: 1) Representation of Trial Data for Predictive Modeling: how to represent the raw sequential data in a way that is most suitable for prediction modeling; 2) Development of Models for Prediction of Trial Outcomes: which model is the most suitable for prediction of outcomes in sequential trials and how to train a prediction model from highly-dimensional multi-therapy recipient sequential data; and 3) Guiding Therapy of a Child with Autism Based on an Early Classification Model: how to adjust and extend the previously developed approach by the project team to guide trials.