Performance Evaluation of SVM with Non-Linear Kernels for EEG-based Dyslexia Detection

Dyslexia is a neurodevelopmental disorder that involves difficulty in interpreting, reading, and writing but does not necessarily affect intelligence. Several behavioral symptoms have been utilized to identify Dyslexic patients. This paper proposes the analysis of EEG signals to diagnose Dyslexic people. Preprocessing the raw EEG data, feature extraction, feature grouping, and transferring particular attributes to a machine learning-based classifier are all part of the implementation. The feature grouping block has received a combination of Statistical, Hjorth, Frequency, and Katz Fractal Dimension attributes. Instead of collecting data from all channels, channels are aggregated and sent to a classifier to determine which part of the brain is engaged for a given activity, resulting in fewer electrodes for Dyslexia detection. In this work, an SVM classifier with non linear kernels is implemented. The performance of the Gaussian (RBF) Kernel, Polynomial Kernel, and Sigmoid Kernel has been evaluated. The Gaussian (RBF) Kernel produces promising results due to its decreased error rates. The deployed SVM model's performance was assessed using both speech and non-speech stimulus. This framework was examined on 391 participants, the most instances evaluated by any other researcher in the development of a feature-based machine learning technique. We attained a maximum accuracy of 62.4 percent for no-speech stimuli using the RBF Kernel, which is significant with the large dataset.