Schedule of Events | Symposia

Sketchpad Series

Preliminary Exploratory Analysis of Autistic Camouflaging using Machine Learning

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

Dan Kingsley¹ (dkingsl2@gmu.edu), Goldie McQuaid¹, Allison Jack¹; ¹George Mason University

Autistic populations have greater mental health challenges such as anxiety than non-autistic populations. Camouflaging, the attempt to appear “non-autistic” in social situations, has been correlated with higher levels of anxiety. We used data from autistic participants in the GENDAAR cohort (NDA data Collection #2021), exploring variable importance in relation to camouflaging (assessed via the Camouflaging Autistic Traits Questionnaire [CAT-Q]) with a machine learning model. Data from a subset of individuals (N=101; 15-30y; 46 assigned female at birth [AFAB]) were preprocessed and run through ridge, LASSO, and elastic net regression machine learning models in R. Features evaluated (n=23) included domain scores from measures of autistic traits (SRS-2), restricted repetitive behaviors (ADI-R:C), and co-occurring psychiatric and behavioral symptoms/traits (ABCL/CBCL). The best fit models were the LASSO (R2 =.05; RMSE=22.67; MAE=17.59) and the elastic net (R2 =.02; RMSE=23.31; MAE=17.78). The variables of most importance were sex assigned at birth (SAB), anxiety problems, and unusual sensory interests, all of which were correlated with higher predicted levels of camouflaging; and rule-breaking, which was correlated with lower levels. These preliminary findings indicate that being AFAB, higher levels of anxiety, and unusual sensory interests predict greater camouflaging. This suggests that AFAB individuals struggle with anxiety more than AMAB (assigned male at birth) individuals partially because of mental burden associated with camouflaging. Rule-breaking behaviors may indicate lower recognition of social rules, underlying factors which may translate into less awareness or usage of camouflaging. Final models will be run with MRI data to explore the neural correlates of camouflaging.

Topic Area: METHODS: Other