Evaluating the Effect of Dropout and L2 Regularization on EEG-based ADHD Detection Using Deep Learning
Adolf Lobowicz, Akriti Srivastava
International Community School, Kumasi, Ghana
The British School of Milan, Milan, Italy
Publication date: November 20, 2025
International Community School, Kumasi, Ghana
The British School of Milan, Milan, Italy
Publication date: November 20, 2025
DOI: http://doi.org/10.34614/JIYRC2025II42
ABSTRACT
ADHD is a neurodevelopmental disorder that affects brain development. Conventional diagnosis relies on behavioural assessments which are often tedious, and subjective. EEG signals, however, record brain activity non-invasively and objectively, offering a promising diagnostic alternative. In this paper, we trained a feedforward neural network using a publicly available EEG dataset from Kaggle to explore the influence of dropout and L2 regularization on overfitting and generalization. Through empirical tests on four unique configurations with five trials each, we found that the combined use of dropout and L2 regularization achieved the best balance between accuracy and reduced overfitting, creating a suitable diagnostic model.
ADHD is a neurodevelopmental disorder that affects brain development. Conventional diagnosis relies on behavioural assessments which are often tedious, and subjective. EEG signals, however, record brain activity non-invasively and objectively, offering a promising diagnostic alternative. In this paper, we trained a feedforward neural network using a publicly available EEG dataset from Kaggle to explore the influence of dropout and L2 regularization on overfitting and generalization. Through empirical tests on four unique configurations with five trials each, we found that the combined use of dropout and L2 regularization achieved the best balance between accuracy and reduced overfitting, creating a suitable diagnostic model.
