As indicated by longitudinal observation, autism has difficulty controlling emotions to a certain extent in early childhood, and most children's emotional and behavioral problems are further aggravated with the growth of age. This study aimed at exploring the correlation between white matter and white matter fiber bundle connectivity characteristics and their emotional regulation ability in children with autism using machine learning methods, which can lay an empirical basis for early clinical intervention of autism. Fifty-five high risk of autism spectrum disorder (HR-ASD) children and 52 typical development (TD) children were selected to complete the skull 3D-T1 structure and diffusion tensor imaging (DTI). The emotional regulation ability of the two groups was compared using the still-face paradigm (SFP). The classification and regression models of white matter characteristics and white matter fiber bundle connections of emotion regulation ability in the HR-ASD group were built based on the machine learning method. The volume of the right amygdala (R = 0.245) and the volume of the right hippocampus (R = 0.197) affected constructive emotion regulation strategies. FA (R = 0.32) and MD (R = 0.34) had the predictive effect on self-stimulating behaviour. White matter fiber bundle connection predicted constructive regulation strategies (positive edging R = 0.333, negative edging R = 0.334) and mother-seeking behaviors (positive edging R = 0.667, negative edging R = 0.363). The emotional regulation ability of HR-ASD children is significantly correlated with the connections of multiple white matter fiber bundles, which is a potential neuro-biomarker of emotional regulation ability.