ImRA_SRNet: Improved Residual Attention-based Simple Recurrent Neural Network-based Autism Detection Using Eye Tracking

This research introduces a novel method for autism detection using eye-gaze analysis. The input data are sourced from publicly available datasets and undergo preprocessing using contrast-limited adaptive histogram equalization to enhance the image quality. Image augmentation techniques are then applied to expand the dataset to eliminate potential biases in the model’s outcomes. For autism detection, an Improved Residual Attention-based Simple Recurrent Neural Network model is proposed, which combines residual learning with attention mechanisms to better capture critical gaze patterns associated with autism. The model’s performance is further optimized through the use of the Enhanced Hippopotamus Optimization algorithm, which refines the loss function during training to make detection more accurate. The proposed method aims to enhance the reliability and precision of autism detection based on eye-gaze data and acquired Accuracy, Specificity, Precision, Recall, F1-Measure, and Mean square error (MSE) values of 98.51, 98.25, 97.32, 98.02, 97.97, and 1.49, respectively.