Automatic detection of attention deficit hyperactivity disorder using machine learning algorithms based on short time Fourier transform and discrete cosine transform.

Objective: Attention deficit hyperactivity disorder (ADHD) is a predominant neurobehavioral illness in minors and adolescents, with overlapping symptoms that complicate established diagnostic approaches. Electroencephalography (EEG) is a noninvasive system for analyzing brain action, with the possibility of automated diagnosis.

Method: This study investigates the use of electroencephalogram decomposition approaches for better detection of ADHD. We used independent component analysis (ICA) to eliminate noise and artifacts of EEG. EEG signals were decomposed into subbands using robust short time Fourier transform (STFT) and discrete cosine transform (DCT) decomposition methods. These sub-bands and EEG signals are input for the machine learning algorithm that could distinguish between healthy volunteers from those having ADHD.

Result: The findings show that STFT techniques perform better than DCT. According to the experiment's results, the STFT method had the highest sensitivity rates. However, combo of Fp1Fp2F3F4P3C3 (6 electrodes placements) achieves 91% accuracy and 90% on Fp1F3C3P3O1 (combination of 5 electrodes) when using STFT-XGBoost. On combination Fp1F3 F7F8 (4 electrodes), the accuracy of Logistic Regression is 89% and 88% for combinations of three electrode placements F3F4C4, F3C3F7, and F3O2F7. Random Forest outperforms with an accuracy of 89% with the classification algorithm on a combination of all (19) electrode placements.

Novelty: This automated detection technology could help clinicians improve early diagnosis and personalized treatment options. The current study's findings contribute to the literature through uniqueness, and the suggested technique can eventually be used as a medical tool for diagnosis in the future.