Table 2 Summary of classification balanced accuracy and sensitivity

	Balanced accuracy (%)				Sensitivity (%)
	SVM	Chance [99.9% CI]	KNN	Chance [99.9% CI]	SVM	Chance [99.9% CI]	KNN	Chance [99.9% CI]
Within-participant classification (# of participants)
Critical nodes (n = 15)	70.4	60.0 [58.4–61.7]	72.8	58.9 [57.4–60.6]	83.7	66.7 [59.5–73.0]	84.4	54.5 [45.5–62.0]
Language error nodes (n = 10)	70.4	61.0 [59.1–63.3]	78.1	59.5 [57.6–61.5]	88.5	69.3 [59.7–77.2]	85.3	53.3 [42.0–62.7]
Speech arrest nodes (n = 8)	72.9	62.7 [60.6–65.1]	77.4	60.6 [58.5–63.2]	89.4	72.7 [64.3–80.5]	87.4	55.2 [42.8–67.0]
Across-participant classification (# of participants)
Critical nodes (n = 16)	65.9	60.7 [59.1–62.7]	66.0	60.7 [59.0–62.7]	76.4	68.0 [62.0–74.5]	77.3	67.6 [60.8–74.0]
Language error nodes (n = 13)	68.9	62.4 [59.9–65.1]	69.5	62.4 [60.1–65.0]	86.7	72.3 [64.7–79.4]	89.5	72.4 [64.7–79.4]
Speech arrest nodes (n = 13)	74.0	65.3 [62.9–67.8]	71.1	65.1 [62.9–67.8]	97.5	76.0 [67.4–85.1]	90.0	75.8 [67.9–83.9]

Median balanced accuracy and sensitivity values for within-participant and across-participant critical, language error, and speech arrest node prediction. Empiric chance intervals for both SVM and KNN are generated by applying trained classifiers with shuffled class labels, and resampling these chance predictions to generate confidence intervals. The estimates presented are average accuracy per participant for the true data and shuffled data. For within-participant classification, participants with at least four nodes of the relevant class were included, and for across-participant classification, participants with at least one node of the relevant class were included. Classification accuracy for all models (balanced accuracy, sensitivity) exceeded the 99.9% confidence interval limit, and are highlighted in bold.

Quick links

Search