Efficient Feature Selection and Machine Learning Based ADHD Detection Using EEG Signal

Abstract

Attention deficit hyperactivity disorder (ADHD) is one of the most common psychiatric and neurobehavioral disorders in children, affecting 11% of children worldwide. This study aimed to propose a machine learning (ML)-based algorithm for discriminating ADHD from healthy children using their electroencephalography (EEG) signals. The study included 61 children with ADHD and 60 healthy children aged 7-12 years. Different morphological and time-domain features were extracted from EEG signals. The t-test (p-value < 0.05) and least absolute shrinkage and selection operator (LASSO) were used to select potential features of children with ADHD and enhance the classification accuracy. The selected potential features were used in four ML-based algorithms, including support vector machine (SVM), k-nearest neighbors, multilayer perceptron (MLP), and logistic regression, to classify ADHD and healthy children. The overall prevalence of boys and girls with ADHD was 48.9% and 56.5%, respectively. The average age of children with ADHD was 9.6±1.8 years. Our results illustrated that the combination of LASSO with SVM classifier achieved the highest accuracy of 94.2%, sensitivity of 93.3%, F1-score of 91.9%, and AUC of 0.964. Our results also illustrated that MLP was the second-best ML-based classifier, which gave 93.4% accuracy, 91.7% sensitivity, 91.1% F1-score, and 0.960 AUC. The findings indicated that the combination of the LASSO-based feature selection method and SVM classifier can be a useful tool for selecting reliable/potential features and classifying ADHD and healthy children. Our proposed ML-based algorithms could be useful for the early diagnosis of children with ADHD.

Optimal Channels and Features Selection Based ADHD Detection From EEG Signal Using Statistical and Machine Learning Techniques

“…Altınkaynak et al [4] also extracted morphological, non-linear, and wavelet features to diagnose children with ADHD. In our current study, we have also extracted time domain, morphological, and non-linear features on the basis of previous studies [4], [24], [25].…”

“…Moreover, the channel selection approach was used as an effective tool by many researchers in different fields, such as EEG emotion [26]- [30], personal identification [31], [32], user identification [33], seizure detection [34], [41], intruder detection [35], screening of alcoholism [36], depression detection [39], [40], detecting drowsiness [42], auditory attention detection [37], [38], brain-computer interfaces [43], [44] and so on. It was noted that several studies proposed effective predictive-based approaches for the detection of children with ADHD without selecting potential channels from EEG signals [4], [12]- [14], [24], [25], [45]- [48]. Although, only a few studies gave more attention on the selection of effective channels [49], [50], From these surveys, we got the motivation to do this work which will be beneficial in the ADHD research domain as well as the clinical domain.…”

“…Feature selection is very important because it removes redundant features, improves the model's performance, and reduces overfitting problems. In the literature, there were various feature selection techniques such as principal component analysis (PCA) [49], [51], minimum redundancy maximum relevance (mRMR) [52], MI [53], t-test [4], [25], support vector machine recursive elimination (SVM-RFE) [52], least absolute shrinkage and selection operator (LASSO) [25], logistic regression (LR) [10], and so on, which were widely used to select the potential features in any domain like ADHD. For example, Tenev et al [54] used the forward selection method to select the significant features for ADHD.…”

“…Chen et al [52] applied the mrMR-based approach to select the most discriminative features. Maniruzzaman et al [25] implemented two effective feature selection methods as independent t-test and LASSO for the identification of potential biomarkers of children with ADHD. In this current study, we also applied LASSO logistic regression (LASSO-LR) based model for the selection of more effective features from the selected channels.…”

“…, ML-based approaches have been widely used in the field of medical imaging [46], time series [20], [21], [55], [56], and bioinformatics [57]. Like other domains, ML-based approaches were also widely used to discriminate children with ADHD from healthy control [4], [10], [13], [24], [25], [47], [49], [52], [58]- [60]. For example, Muller et al [61] proposed SVM-based children with ADHD detection system.…”