A Predictive Multimodal Framework to Alert Caregivers of Problem Behaviors for Children with ASD (PreMAC)

Area of Science:

• Neurodevelopmental Disorders
• Artificial Intelligence in Healthcare
• Behavioral Science

Background:

• Autism Spectrum Disorder (ASD) affects 1 in 54 US children, with two-thirds exhibiting problem behaviors.
• Predicting and managing problem behaviors is crucial for caregivers of children with ASD.
• Current human prediction methods have limitations, necessitating technological assistance.

Purpose of the Study:

• To propose PreMAC, a machine learning framework for predicting problem behaviors in children with ASD.
• To introduce the M2P3 platform for collecting multimodal data to train PreMAC.
• To assess the feasibility and accuracy of the M2P3 platform and PreMAC.

Main Methods:

• Development of the PreMAC predictive framework using machine learning.
• Design of the M2P3 multimodal data capture platform.
• Integration of the interview-informed synthesized contingency analysis (IISCA) for data collection.
• Feasibility study with seven children (ages 4-15) with ASD.

Main Results:

• The M2P3 platform was well-tolerated by children with ASD.
• PreMAC demonstrated high prediction accuracy for precursors of problem behaviors.
• The framework shows promise for real-time alerts to caregivers.

Conclusions:

• The PreMAC framework and M2P3 platform are feasible for use with children with ASD.
• Machine learning can effectively predict problem behaviors in ASD.
• This technology offers a novel approach to support caregivers in managing challenging behaviors.