An ensemble-based Machine learning technique for dyslexia detection during a visual continuous performance task

Background: Dyslexia is a neurological disorder which affects the learning of individuals suffering from it, especially children. It causes reading and writing difficulties, leading to anger, frustration, poor self-esteem, and other negative emotions. Early diagnosis of dyslexia can be extremely useful for dyslexic children since their learning requirements can be correctly handled.
Method: This study aimed to propose an efficient method for dyslexia detection via evoked-related potentials (ERPs) during a Visual Continuous Performance Task (VCPT). To this end, after extracting the latency and amplitude of the event-related potential components and a statistical analysis between the two groups, the number of data dimensions is reduced by the KS-test and principal component analysis (PCA). This method used the ensemble learning classification technique to classify the normal and dyslexic groups, including five classifiers, namely a Support Vector Machine (SVM), Decision Tree, Linear Discriminant Analysis (LDA), K-nearest Neighbors (KNN), and Naive Bayesian. Moreover, we have used an 8-fold cross-validation method to assess how generalized the process is and, more importantly, to have a well-performed control on overfitting.
Results: According to this approach, the overall average classification accuracy is 87.5%, and the sensitivity and specificity are 81.2% and 93.7%, respectively.
Conclusion: The suggested technique can improve children's education before they start school and have any skills in writing and reading. The most discriminative electrodes were found in the left hemisphere, which can be related to the Wernicke and Broca areas that have an essential role in the brain's reading, writing, and language-related functions.