Machine Learning-Based Detection of Autism Spectrum Disorder Using Attention Mechanisms in Eye-Tracking Data

Effective and accurate diagnosis of Autism Spectrum Disorder (ASD) has become crucial for early intervention and a better outcome for patients. This study attempts to improve ASD detection through machine learning (ML) techniques applied to eye-tracking data. We developed a machine learning model that utilizes eye movement sequences, representing the participant's attention mechanisms, to enhance the accuracy of ASD detection. We trained our model, consisting of multiple GRUs and an ANN architecture, on a dataset of eye-tracking data without augmentation. The model was trained on sequences of distances between eye fixations, transformed sequences, and durations of fixations and saccades. To evaluate the performance of our model and compare it with other works, we employed a validation dataset. Our ML model demonstrated superior performance in ASD detection compared to other works on the same dataset. Our model achieved higher accuracy, precision, and recall. By incorporating sequence-based features, our model effectively distinguished between individuals with ASD and typically developing individuals, achieving an impressive precision of 90.68% on the validation dataset. Our findings highlight the potential of a temporal analysis for improved ASD detection and provide a promising direction for further advancements in the field of eye-tracking-based diagnosis and intervention for neurodevelopmental disorders.