Table 5.
Applications of machine learning methods
| References | Data type | Subjects | Method | Disease/state | Application | Effect evaluation |
|---|---|---|---|---|---|---|
| Qaraqe et al. [162] | EEG | CHB-MIT dataset | CSP | Epilepsy | Utilized the CSP approach for seizure detection | A sensitivity of 100%, a detection latency of 7.28 s, and a false alarm rate of 1.2 per hour were successfully attained in this article |
| Dissanayake et al. [163] | EEG | CHB-MIT dataset | CSP | Epilepsy | Adopted the CSP algorithm for patient-independent seizure prediction | Accuracy achievements of 88.81% and 91.54% were reported in this article |
| Liu et al. [164] | EEG | BCI competition III-4a BCI competition IV-2a strokes (n = 5) | CSP | Stroke | Investigated the rehabilitation of stroke patients using the CSP algorithm | High accuracies were achieved in comparison with seven state-of-the-art approaches, as highlighted in this article |
| Alturki et al. [165] | EEG |
Normal (males = 10) ASD (males = 6, females = 3) CHB-MIT dataset |
CSP | Epilepsy and ASD | Applied the CSP algorithm for the diagnosis of epilepsy and autism | Accuracy rates of approximately 98.46% for diagnosing ASD and 98.62% for epilepsy were achieved in this article |
| Jamal et al. [166] | EEG |
ASD (n = 12) Normal (n = 12) |
LDA | ASD | Carried out LDA to classify ASD | Leave-one-out cross-validation of the classification algorithm resulted in a best performance of 94.7% accuracy, with corresponding sensitivity and specificity values of 85.7% and 100%, as reported in this article |
| Jeong et al. [167] | EEG |
PDD (n = 26) AD (n = 26) Normal (n = 26) |
LDA | PDD and AD | Applied LDA to distinguish between PD-related dementia and AD | A maximum performance of 80.19% accuracy was achieved using LDA with WC in this article |
| Boostani et al. [168] | EEG |
SZ (males = 13) Normal (males = 18) |
LDA | SZ | Adopted LDA in the diagnosis of SZ | Accuracies of 87.51%, 85.36%, and 85.41% were achieved for BDLDA, LDA, and Adaboost, respectively, in this article |
| Rajaguru et al. [169] | EEG | Epilepsy (n = 20) | LDA | Epilepsy | Used the LDA approach for the classification of epilepsy | When the dB2 and dB4 wavelets were classified with LDA, average classification accuracies of 95.83% and 95.03% were obtained, as claimed in this article |
| Kang et al. [170] | EEG |
ASD (boys = 39, girls = 10) TD (boys = 36, girls = 12) |
SVM | ASD | Employed the SVM method to identify children with ASD | Combining two types of data resulted in a maximum accuracy of 85.44%, with AUC = 0.93 when 32 features were selected in this article |
| Fu et al. [171] | EEG | Bonn dataset | SVM | Epilepsy | Adopted the SVM approach for the classification of epilepsy | A 99.125% accuracy of the algorithm with the theta rhythm of EEG signals was achieved in this article |
| Shen et al. [172] | EEG | Normal (n = 10) | SVM | Mental fatigue measurement | Utilized the SVM method for mental fatigue measurement | An accuracy of 87.2% for the probabilistic multi-class SVM compared to 85.4% using the standard multi-class SVM was reported. With confidence estimates aggregation, the accuracy increased to 91.2% |
| Liu et al. [173] | iEEG | Epilepsy (n = 21) | SVM | Epilepsy | Performed seizure detection in long-term EEG using the SVM method | A sensitivity of 94.46%, specificity of 95.26%, and a false detection rate of 0.58/h for seizure detection in long-term iEEG were achieved in this article |
| Zhou et al. [174] | EEG | CHB-MIT dataset | CNN | Epilepsy | Detected seizures through CNN models | The article achieved a convincing performance with an accuracy of 94.67% on the test data |
| Hassan et al. [175] | EEG | Bonn dataset | CNN | Epilepsy | Detected epilepsy through the 1D-CNN approach | Using frequency domain signals, average accuracies of 96.7%, 95.4%, and 92.3% for the three experiments were achieved in the Freiburg database, while average accuracies for detection in the CHB-MIT database were 95.6%, 97.5%, and 93% for the three experiments in this article |
| Hassan et al. [176] | EEG |
SZ (n = 14) Normal (n = 14) |
CNN | SZ | Detected SZ through the 1D-CNN approach | The article effectively predicted two, three, four, and five classes with accuracies of 100%, 99%, 94.6%, and 94%, respectively, for the Bonn dataset and 98% for the CHB-MIT dataset |
| Dong et al. [177] | EEG |
ASD (children = 86) Normal (children = 89) |
CNN | ASD | Applied the CNN method for the assessment of ASD in children | Accuracies of 90% and 98% were achieved for subject-based and non-subject-based testing, respectively, in this article |
| Aliyu et al. [178] | EEG | Bonn dataset | CNN | Epilepsy | Detected epileptic EEG signals using CNN | The method was claimed to outperform its counterparts, achieving individual/sample accuracy of 92.63%/83.23%, as reported in this article |
| Lee et al. [179] | EEG |
PD (n = 20) Normal (n = 20) |
RNN | PD | Combined CNN with RNN for the identification of PD | An accuracy of 99.2%, precision of 98.9%, and recall of 99.4% in differentiating PD from healthy controls were achieved in this article |
| Sarkar et al. [180] | EEG | Normal (male = 1, female = 1) | RNN | Mental depression | Detected mental depression through RNN | The article achieved the highest accuracies of 97.50% in the training set and 96.50% in the test set |
| Mishra et al. [181] | EEG | Sleep-EDF dataset | RNN | Sleep stages | Employed CNN and RNN for sleep stage classification | Efficient classification performance in sleep stage N1, as well as improvement in subsequent stages of sleep, was reported in this article |
| Michielli et al. [182] | EEG | Normal (n = 10) | LSTM | Sleep stages | Used LSTM for the classification of different sleep stages | The overall percentage of correct classifications for five sleep stages was found to be 86.7% in this article |
| Hu et al. [183] | EEG | CHB-MIT dataset | LSTM | Epilepsy | Established LSTM models to achieve the automatic detection of epilepsy | A mean sensitivity of 93.61% and a mean specificity of 91.85% were achieved on a long-term scalp EEG database in this article |
| Koya et al. [184] | EEG | Normal (n = 10) | LSTM | Emotion | Adopted LSTM to recognize and classify different emotions | In this article, the LSTM + CNN model outperformed traditional or deep learning models, achieving an accuracy of 64% |
| Lee et al. [185] | EEG | Normal (n = 10) | LSTM | Sleep stages | Detected drowsiness indicators using the LSTM method | The LSTM-CNN model in this article demonstrated an average accuracy of 85.6% and a kappa index of 0.77 for the three-class classification problem |
AD Alzheimer’s disease, ASD autism spectrum disorders, BDLDA block diagonal LDA linear discriminant analysis, CHB-MIT Children’s Hospital Boston (CHB) and the Massachusetts Institute of Technology, CNN convolutional neural network, CSP common spatial patterns, EEG electroencephalography, iEEG intracranial electroencephalography, LDA linear discriminant analysis, LSTM long short-term memory, PD Parkinson’s disease, PDD Parkinson’s disease-related dementia, RNN recurrent neural network, SVM support vector machine, SZ schizophrenia, BCI brain-computer interface, WC wavelet coherence, TD typically developing, EDF European Data Format