Treffer: Optimizing a Convolutional Neural Network Model in Amazon SageMaker for an Autism Detection Tool, EZ Autism Screener.

Autism is a neurological and developmental disability caused by changes in the brain's development that also affects the facial tissues. Thus, children with autism show distinct facial features that are not present in average children. Studies reveal increasing prevalence of autism; however, acquiring affordable, trouble-free, and practical early screening tools is a current concern. This impacted early detection and diagnosis of autism which also influenced effective intervention. How can we employ innovative technology, like computer vision and deep learning to build an inexpensive and universally accessible autism screener to prevent late detection and diagnosis of autism? How can we enhance public access to this screening tool and minimize the difficulties involved in the assessment process? We built a basic Convolutional Neural Network (CNN) binary image classifier with seven (7) layers including the input and output layers. This initial model produced positive outcomes with a specificity score of 90.38% and this is the most important evaluation metric for health-related problems like screening for autism. We optimized this model by performing hyperparameter tuning using a cloud machine learning platform, Amazon SageMaker. The tuning job also produced a superior and robust model as reflected in the F1 score of 94.74%. It correctly classified 95% of the images. The model's specificity indicates it correctly identified 100% of those without autism as non-autistic; the recall indicates it correctly identified 90% of those with autism as autistic while its precision indicates a 100% probability that those identified by the model as autistic have autism. Tuning this model took 6 minutes. We integrated this model into a simple iOS application for mobile devices. [ABSTRACT FROM AUTHOR]

Copyright of Journal of Information Systems Applied Research is the property of Information Systems & Computing Academic Professionals (ISCAP) and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)