Enhanced Eeg Signal Classification for Autism Spectrum Disorder Using Riemannian Geometry and Ensemble Learning Techniques

Autism spectrum disorder (ASD) poses a severe challenge to effective communication and social interaction abilities for a large number of individuals worldwide. The P300 signal is difficult to detect in individuals with ASD due to noise, low amplitude, and increased latency compared to others. We enhance in this work the P300 classification in electroencephalogram (EEG) signals for autism disease using Riemannian geometry along with various conventional classifiers and ensemble learning approaches like bagging and boosting techniques (AdaBoostM1, GentleBoost, and LogitBoost), and developing sophisticated pre-processing methods for feature extraction. Using the BCIAUT-P300 dataset, our work achieved 96.37% accuracy, 97.51% sensitivity, and 91.11% specificity compared to existing processes, ranging from 67.2% to 92.3% accuracy. This work highlights the potential of our technique in assisting in diagnosis and supportive ASD technologies