A Hybrid Approach to Attention Deficit Hyperactivity Disorder Detection Leveraging Transformer and XGBoost Models...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Diagnostics

Background:

• Early detection of neurodevelopmental disorders like ADHD is vital for effective intervention.
• Traditional diagnostic methods face challenges including subjectivity, resource limitations, and diagnostic biases, leading to potential under/overdiagnosis.
• Accurate diagnosis supports social, cognitive, and mental development.

Purpose of the Study:

• To develop an advanced ensemble model for enhanced accuracy and efficiency in diagnosing neurodevelopmental disorders.
• To address the limitations of traditional diagnostic approaches through innovative computational methods.

Main Methods:

• Proposed an ensemble model, XSparseFormerNet, integrating a custom encoder-decoder Transformer with attention mechanisms and an XGBoost model.
• Utilized a preprocessed electroencephalogram (EEG) dataset for training and validation.
• Employed a custom transformer architecture combined with the XGBoost gradient boosting algorithm.

Main Results:

• The XSparseFormerNet model achieved 85% accuracy in diagnosing neurodevelopmental disorders.
• The proposed model demonstrated superior performance across various evaluation metrics compared to traditional methods.
• The study highlights the potential of ensemble models in improving diagnostic accuracy for neurological conditions.

Conclusions:

• The XSparseFormerNet model offers a significant advancement in the early detection of neurodevelopmental disorders.
• This research provides a robust methodology for future studies in disorder detection using AI and EEG data.
• The findings underscore the importance of leveraging advanced AI techniques for more precise and efficient medical diagnostics.