Schedule of Events | Symposia

Reducing Bias in Autism Spectrum Disorder Diagnostic Procedures Through Machine Learning

Poster Session F - Tuesday, April 1, 2025, 8:00 – 10:00 am EDT, Back Bay Ballroom/Republic Ballroom

Georgia Agoritsas¹ (gagoritsas@outlook.com); ¹The Bronx High School of Science

Autism spectrum disorder (ASD) is a developmental disability that causes one to act, think, and communicate differently than what is considered the social norm. Historically, ASD was associated with males, a concept rooted in Asperger’s syndrome and its original studies. Recently, more females have been diagnosed, but many still remain undiagnosed. Therefore, this project examined correlations between ASD diagnosis and traits, such as sex, ethnicity, or score on one of ten questions from the AQ-10 autism diagnostic assessment. Using a UCI Adult Autism Screening dataset of around 700 participants, machine learning was incorporated to analyze the dataset using data preprocessing, correlation analysis, and predictive modeling. A correlation map determined a significant relationship between AQ-10 responses and diagnosis (correlation ≥ 0.54) and a negative correlation between gender and diagnosis (-0.12). This suggests that certain autism traits can be focused on more than others when being diagnosed with autism, and gender does not significantly impact an autism diagnosis. Altogether, this supports the idea that anyone can be diagnosed with autism if they exhibit the proper criteria for it and that a bias should not be held when diagnosing a patient with autism. Machine learning could further reduce this bias in autism diagnoses, providing a fairer, more accurate assessment process.

Topic Area: METHODS: Other