Supervised Machine Learning for a Novel Autism Prediction Tool in Adults

For many late-identified autistic adults, the realization that they are autistic can be a key first step to accessing supportive resources and accommodations and developing self-understanding. We introduce a novel screening tool for autistic traits designed primarily by autistic adults. It includes options to assess masking, sensory processing differences, commonly co-occurring medical or mental health conditions, and questions about social and communication differences and repetitive behavior. We used a simple supervised machine learning algorithm to generate a score predicting whether the individual is autistic. Our results indicate that this algorithm was able to distinguish respondents who are non-autistic from respondents who self-reported being formally diagnosed as autistic. Additionally, our algorithm remained effective in identifying autism in respondents with other under-represented identities. Finally, we separately analyzed the responses of self-identified autistic individuals and found a high degree of overlap in the responses of self-identified and formally diagnosed autistic respondents.