Abstract
To uncover neurobiological indicators related to autism spectrum disorders and developmental dyslexia, this article gives a full overview of the most recent advances in machine learning and deep learning methods based on electroencephalography. We look into methodological pipelines that include signal gathering, preprocessing, feature engineering, model selection, and interpretability procedures. We based these pipelines on 15 peer-reviewed research papers published between 2013 and 2025. Most of the research employed the 10-20 system for resting-state EEG and followed MATLAB, MNE-Python, or EEGLAB guidelines for preprocessing. The feature sets included spectral power, functional connectivity, task-evoked potentials, and entropy measures. People used many standard ML methods, such as support vector machines and random forests, as well as more advanced models, like deep neural networks and transformer-based architectures. Several studies found that both dyslexic and ASD groups did well at classifying, with accuracy scores between 82% and 99.2%. The new models could be used in therapeutic settings, but there are still problems with how easy they are to understand and how well they apply to a wide range of situations. This is especially true for ASD because its spectrum is so varied.