Multimodal Behavioral Signal Fusion Using Wearable IoT and Context-Aware Machine Learning for Early Autism Escalation Detection

Abstract

Behavioral escalation in children with autism spectrum disorder is a multifactorial phenomenon influenced by physiological, behavioral, and environmental factors that interact dynamically over time. Early detection of escalation remains challenging because many existing systems rely on single-modality signals or static thresholds that fail to capture individualized and context-dependent patterns. This study proposes a multimodal behavioral signal fusion framework that integrates wearable Internet of Things sensing with context-aware machine learning to enable early and reliable escalation detection. The framework combines physiological indicators, motion dynamics, and environmental context using adaptive feature weighting and temporal modeling to capture personalized escalation signatures. Experimental evaluation using simulated autism care scenarios informed by empirical studies demonstrates that multimodal fusion significantly improves early detection accuracy and robustness compared to unimodal approaches while reducing false alerts. The proposed architecture advances personalized, privacy-aware autism care by leveraging context-aware intelligence for proactive intervention.