Autism Spectrum Disorder (ASD) is a heterogeneous neurodevelopmental disorder (NDD) with a high rate of comorbidity. The implementation of eye-tracking methodologies has informed behavioral and neurophysiological patterns of visual processing across ASD and comorbid NDDs. In this study, we propose a machine learning method to predict measures of two core ASD characteristics: impaired social interactions and communication, and restricted, repetitive, and stereotyped behaviors and interests. Our method extracts behavioral features from task performance and eye-tracking data collected during a facial emotion recognition paradigm. We achieved high regression accuracy using a Random Forest regressor trained to predict scores on the SRS-2 and RBS-R assessments; this approach may serve as a classifier for ASD diagnosis.