A deep-learning approach that predicts young children's later externalizing behavior problems from the moment-to-moment dynamics of mother-child interaction during a challenging task. The Attention-based Sequential Behavior Interaction Modeling (ASBIM) model encodes the sequence of dyadic behaviors with an attention mechanism and adds a child's inhibitory control as a personalized feature, improving on a residual dynamic structural equation baseline. Students learn how temporal interaction patterns, especially maternal autonomy support following child defeat, carry predictive signal for developmental outcomes.
Objective: Predicting children's future levels of externalizing problems helps to identify children at risk and guide targeted prevention. Existing studies have shown that mothers providing support i…