LAUSR

Screening and timely diagnosis of children with attention-deficit/hyperactivity disorder (ADHD) are currently an important challenge in studies of children’s ethology for increasing community mental health. Nevertheless, most studies on ADHD employed self-report questionnaires in which the reliability is questionable. A few of the studies for diagnosing the ADHD, which used the EEG signals, was not able to enhance the accuracy due to the use of stand-alone classifiers. Therefore, in this paper, we provided a novel analysis framework by combining a set of fuzzy inference system (FIS), called the mixture of expert fuzzy models (MEFM), which have high potential for diagnosing the ADHD children. Within this framework, we developed the MEFM utilizing features extracted from the continuous wavelet transform scalograms of EEG signals obtained from five different channels (Fz, Cz, Pz, C3 and C4). Then, we tuned the internal parameters of the MEFM employing the genetic algorithm. We, also, evaluated the MEFM classifier applying the tenfold cross-validation method and compared this classifier with stand-alone FIS and support vector machine (SVM). Experimental results of MEFM classifier due to collective decision-making of the gating network exhibited a significant improvement (99.01% accuracy) compare to stand-alone classifiers (98.07, 97.81% and 88.25 for RBF-SVM, stand-alone fuzzy inference system (SFIS), linear-SVM). These results also illustrate that the MEFM classifier (due to distributed structure in comparison with the SFIS classifier) extremely reduced the number of fuzzy rules. Therefore, this new fuzzy combination (among the classifiers developed by several EEG channel) not only is a novelty in the fuzzy theory, but also considers being a better alternative for identifying and screening children with ADHD.