Significance of Pre-Processing Phase and Dimensionality Reduction in EEG-based Dyslexia Diagnosis with Novel Features

The specific learning issue known as dyslexia has its roots in the neuroscience of a person. Accurate and fluent word recognition, spelling mistakes, and poor deciphering ability are all symptoms of this disorder. When a student struggles to read and write because of dyslexia, it can have a significant influence on their academic success. It is impossible to overestimate the significance of early identification of learning disorders such as dyslexia. In this study, 53 students were given an N-Back assignment to test their working memory, and EEG recordings were taken of them while they were working on the task. There are 29 people with dyslexia and 24 people with normal literacy. In this work, the importance of pre-processing is demonstrated. In addition to statistical and spectral features, Lempel-Ziv Complexity based features and modified center symmetric Local Binary Pattern for 1D-based features are extracted. Then, features are transferred to the Principal Component Analysis (PCA) block to reduce dimensionality. In order to assess the overall effectiveness of the classifier, a ‘stratified’ 10 fold cross validation approach was utilised. With the MLP kernel, pre-processing steps, and features minimized from PCA block, we were able to achieve a mean correct classification rate of 70.33%, which outperforms current state-of-the-art approaches.