EEG-Based ADHD Diagnosis Using Autoencoder and Reptile Search Algorithm Integrated with Machine Learning

Abstract

Attention Deficit Hyperactivity Disorder (ADHD) is a common neurodevelopmental disorder affecting cognitive and behavioral functions, resulting in ongoing inattention, hyperactivity, and impulsivity. Early and accurate diagnosis is essential, but traditional methods mainly depend on questionnaire-based assessments, detailed interviews with individuals and their families, and reviews of medical history. These are then scored using standardized scales like the Conners Rating Scale, Vanderbilt ADHD Diagnostic Parent Rating Scale, and Adult ADHD Self-Report Scale. However, these methods are often subjective, time-consuming, and costly, which limits their usefulness for early diagnosis. The proposed approach seeks to improve ADHD diagnosis by using machine learning techniques applied to electroencephalogram (EEG) data. Two classifiers, Random Forest and AdaBoost, are used to identify complex patterns in EEG data. Feature selection is performed with the Reptile Search Algorithm combined with an autoencoder for feature extraction, which improves data representation and model accuracy. The performance of this approach is evaluated based on accuracy, precision, recall, F1-score, AUC, and statistical significance at a 95% confidence level. Random Forest outperformed AdaBoost, achieving 92.36% in precision, recall, accuracy, and F1-score, while AdaBoost reached 89.78% in these metrics. Random Forest showed better effectiveness than AdaBoost in distinguishing ADHD cases, with an ROC AUC score of 0.93 and higher diagnostic accuracy. The study demonstrates that machine learning offers a promising, objective, and reliable tool for diagnosis, providing effective alternatives to traditional ADHD assessments for timely intervention and improved treatment management.