Adult Autism Screening

Is this trustworthy?

The purpose of this tool is to provide initial guidance but should not replace professional diagnosis. The model is a simple logistic regression model trained using SK-learn's default params. It does about as well as a more advanced (convolutional) neural network, which seem to be the state-of-the-art according to this paper. On the test set (n=121), it had 98.3% accuracy, and 98.0% F1-score. It's trained on this dataset from the UCI Machine Learning Repository.

Is this safe?

Yes. Unlike other online tests, this one doesn’t gather and sell your data. Don’t trust me? You can read the source code.

Why does this exist?

1. I was curious about the probability of me having ASD, and there didn’t seem to be any publicly available models using ML or Bayes Theorem, so I made my own. Due to the nature of ASD, it’s quite hard to self-diagnose. There’s no good data on base rates for autistic symptoms, so simply using Bayes Theorem isn’t as reliable. Therefore, I concluded the best option is to use ML. Some actual researchers talk about this, so it’s hopefully not too outlandish.
2. I wanted to try out a JavaScript framework, which is why this site exists. For me, this site isn’t useful because I trust the other models more. But they agree about 99% of the time anyway.