ChEMA: exploring multimodal digital phenotyping for just-in-time caregiver-centric intervention in childhood challenging behaviours

Challenging behaviours are common in childhood and may include outbursts, aggression, and self-stimulatory behaviours that cause distress or interfere with daily functioning. While evidence-based interventions can be effective, caregivers may find it difficult to accurately characterise behavioural manifestations and contextual factors due to busy schedules, limited expertise, and recall biases. We present ChEMA (Caregiver-centric Hybrid Ecological Momentary Assessment), an AI-augmented system that provides a just-in-time assessment and intervention framework for home use. ChEMA integrates continuous passive sensing of physiological and behavioural cues with lightweight in-the-moment active prompts. This hybrid EMA approach captures rich context in real time while fostering a caregiver–AI feedback loop for timely, adaptive interventions. In a six-week field study involving a ChEMA group (38 families) and a treatment-as-usual (TAU) group (22 families), ChEMA's multimodal behaviour detection model achieved 87% F1. Compared with the TAU group, caregivers in the ChEMA group showed larger reductions in parenting stress and the severity of children's challenging behaviours over time. Qualitative feedback from the ChEMA group confirmed these findings, highlighting timely interventions and low perceived burden. We discuss design methodology and key considerations for home-based, caregiver-centred AI interventions supporting children with social-emotional and behavioural difficulties.