Research Information System
TURK PSIKOLOJI DERGISI, 2024 (SSCI)
Individuals with autism often face challenges in directing visual attention to human faces in natural social situations and in obatining important social cues from facial gestures and emotional expressions. Given these limitations, our study sought to employ machine learning algorithms to differentiate children with autism from their typically developing (TD) peers. Therefore, videos displaying happy, sad, and neutral emotions were created. An eye-tracking device was used to create separate data sets for each emotional state by recording the eye movements of the participants with autism and TD children, aged 18 to 36 months, as they watched videos reflecting emotional states on the screen. Feature selection methods were used to identify discriminative features from the areas highlighted in the videos. Decision Tree, Naive Bayes and K-Nearest Neighbour classification algorithms were then applied to these features. Among the machine learning algorithms used to discriminate features, the K Nearest Neighbour algorithm showed superior performance on neutral emotion videos. The study achieved an 81.45% success rate in differentiating children with autism from their TD peers. Study findings offer promising insights into the potential use of software, powered by machine learning algorithms, in future clinical assessments of autism symptoms.