Table 5.

Comparison of the classification accuracies of fNIRS-based ML pain classification studies with the presented work.

Study	Task	Used features	Classifier	Performance
Lopez-Martinez et al.4	Electrical pain (high/low intensity)	Scalogram-based features (mean, maximum, and std. dev)	Hiearchical Bayesian logistic regression	Accuracy: 81%
Fernandez-Rojas et al.5	Temperature level (cold and hot)–based pain intensity (low and high)—four classes	25 features (time, frequency, and wavelet)	Gaussian SVM, KNN, and LDA	Gaussian SVM accuracy: 94.17%
Fernandez-Rojas et al.6	Temperature level (cold and hot)–based pain intensity (low and high)—four classes	Raw HbO and Hb	Bidirectional LSTM	Accuracy: 90.6%
Khan et al.7	Electrical pain (high/low intensity)	Log energy, crest factor, shape factor, impulse factor, margin factor, mobility, complexity, mean absolute deviation of the first difference, range, and variation in the first difference of HbO and Hb	SVM	Accuracy: 68.51%
Presented work	Electrical pain (high/low intensity)	Preprocessed epochs	CNN	Accuracy: 97%