Developing a Machine Learning Program to Detect Speech Patterns of Autistic Children

Overview

1 in 68 Americans (over 3.5 million people in the United States) is currently affected with autism spectrum disorder (ASD), and more are diagnosed each day. The average age of diagnosis for children with autism is over four years, but distinct signs develop as early as 18-24 months. Early detection comes with a host of benefits (including earlier access to ASD-appropriate education and enhanced parent-child relationships) that ultimately lead to increased emotional, physical, and mental development. Unfortunately, diagnosis is not always easy and many times children and families lose access to this potentially hugely beneficial care. This project works towards a solution for this problem by identifying speech patterns unique to children with ASD. Developed in the Python programming language, this program was trained on recorded transcripts of conversations involving autistic children. A variety of indicators (e.g. usage of different parts of speech) were analyzed during the detection of these patterns. It was observed that children with ASD give significantly more importance to frequently used words and use adjectives, singular proper nouns, base-form verbs, and prepositions significantly more than their counterparts without ASD. In addition, a predictive model was developed to identify ASD from inputted text with an 83% accuracy rate. By developing this program, another step is taken towards a mechanism that can alert caregivers to signs of autism and potentially cut many months off of the age of diagnosis, allowing many to receive the care and consideration that they need sooner.

Competition history

• AJAS 2019 Category not listed

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

• All Behavioural and Social Sciences projects
• AJAS Behavioural and Social Sciences projects
• Every AJAS 2019 project