| Ref | Modality signal | Features | Classification | Emotions | Accuracy (%) |
| [11] | EEG | Energy, entropy | SVM, KNN | Arousal, valence | 86 | [13] | EEG | Min, Max peak, power | LSTM | Arousal, valence, and liking | 87 | [14] | EEG | Min, Max peak, power | ANN | Stress, normal | 60 | [20] | EMG, ECG, RSP | Statistical, energy, entropy | LDA | Joy, anger, sadness, and pleasure | 95 | [24] | BVP, EMG, EDA, RSP | Statistical features | SVM, Fisher LDA | Amusement, contentment, disgust, fear, sadness, and neutral | 92 | [26] | EMG, EDA, ECG | No specific feature | No specific classifier | Arousal, valence | NA | [27] | EEG | Statistical features | SVM, ANN | Positive, negative, and neutral | 91 | [33] | EEG, EMG, Temp, GSR, RSP | Different features | MESAE | Arousal, valence | 77 | [34] | EEG | No specific features | LDA | Arousal, valence | 87 | [35] | EEG | DE, PSD | SVM | Negative, positive, and neutral | 91.5 | [36] | EEG | Spatial, spectral, temporal | CNN | Depression | 86 | [37] | EDA, HR, EMG | No specific features | HMM | Arousal, valence | 81 | [38] | EEG | Average PSD, mean, variance, Shannon’s entropy, zero crossing | LSSVM | Joy, peace, anger, and depression | 65 | [39] | EDA, HR | No specific features | Fuzzy logic | Stress | 99.5 | [40] | EEG | No specific features | Correlation analysis | Neutral, anger, sadness, anxiety, disgust, and surprise | 90 |
|
|
Nomenclature for signal modalities: RSP denotes relative spectral power, EEG denotes electroencephalogram, ECG denotes electrocardiogram, GSR denotes galvanic skin response, EDA denotes electrodermal activity, BVP denotes blood volume pulse, HR/HP denotes heart rate/pulse, and Temp denotes temperature. Nomenclature for classifiers: LDA denotes latent discriminant analysis, KNN denotes K-nearest neighbors, ANN denotes artificial neural network, SVM denotes support vector machine, HMM denotes hidden Markov model, LSTM denotes long-short-term memory, DFA denotes deterministic finite automata, MESAE denotes multiple fusion layer based-ensemble classifier of stacked autoencoder, and MEMD denotes multiencoder to multidecoder.
|