2017/03/15 by Halim Abbas, F. Garberson, Abbas, Halim +5 · 1 citation
Neuroscience · Social Sciences · Biochemistry, Genetics and Molecular Biology · #Autism Spectrum Disorder Research #Child Development and Digital Technology #Genetics and Neurodevelopmental Disorders
paper · pdf · doi:10.48550/arxiv.1703.06076
Existing screening tools for early detection of autism are expensive,\ncumbersome, time-intensive, and sometimes fall short in predictive value. In\nthis work, we apply Machine Learning (ML) to gold standard clinical data\nobtained across thousands of children at risk for autism spectrum disorders to\ncreate a low-cost, quick, and easy to apply autism screening tool that performs\nas well or better than most widely used standardized instruments. This new tool\ncombines two screening methods into a single assessment, one based on short,\nstructured parent-report questionnaires and the other on tagging key behaviors\nfrom short, semi-structured home videos of children. To overcome the scarcity,\nsparsity, and imbalance of training data, we apply creative feature selection,\nfeature engineering, and novel feature encoding techniques. We allow for\ninconclusive determination where appropriate in order to boost screening\naccuracy when conclusive. We demonstrate a significant accuracy improvement\nover standard screening tools in a clinical study sample of 162 children.\n