Autism spectrum disorder (ASD) is linked to atypical sleep, motor, and circadian patterns, suggesting that continuous behavioral monitoring may yield insights into its biological underpinnings. It was examined whether actigraphy-derived features of sleep, activity, and light exposure can predict ASD status and autistic trait severity. Data from 325 autistic and 130 non-autistic participants, collected via GENEActiv accelerometers over three weeks, were processed using GGIR, Chronosapiens, and Accelerometer software to extract 308 features. Exploratory analyses revealed associations between light exposure and ASD traits, while hierarchical clustering identified distinct circadian–behavioral subgroups. Using automated machine learning and Elastic Net logistic regression with cross-validation, we evaluated classification and regression models across feature domains. Models incorporating sleep and activity features achieved the highest predictive accuracy, while light features contributed modestly but meaningfully. These findings demonstrate that wearable-based measures can reliably predict ASD-related traits, supporting their potential for early detection and intervention. Future work will refine light-based features, control for seasonality, and explore causal relationships among behavioral and environmental factors.