Table 2.

Offline performances

Offline performances
Participant	Day	P (correct | f⋆)(%)	p-value
1	1	71.9 ± 2.2	6.29×10−6
	2	89.4 ± 1.2	1.53×10−17
	3	83.9 ± 2.0	1.30×10−12
	4	84.0 ± 1.9	1.30×10−12
	5	82.2 ± 1.7	6.55×10−12
	Avg.	82.3 ± 6.4	1.26×10−6
2	1	62.2 ± 1.8	6.00×10−3
	2	62.0 ± 1.8	1.05×10−2
	3	60.5 ± 2.0	1.76×10−2
	4	91.6 ± 1.7	1.60×10−19
	5	82.5 ± 1.6	6.55×10−12
	Avg.	71.8 ± 14.3	6.82×10−3
3	1	90.3 ± 1.3	1.66×10−18
	2	83.9 ± 1.1	1.30×10−12
	3	72.8 ± 2.9	2.35×10−6
	4	81.0 ± 2.1	1.35×10−10
	5	83.3 ± 1.2	1.30×10−12
	Avg.	82.3 ± 6.3	4.69×10−7
4	1	74.7 ± 1.7	2.82×10−7
	2	92.3 ± 1.6	1.36×10−20
	3	81.5 ± 1.3	3.07×10−11
	4	80.5 ± 2.0	1.35×10−10
	5	83.5 ± 2.2	1.30×10−12
	Avg.	82.5 ± 6.4	5.64×10−8
5	1	82.7 ± 1.3	6.55×10−12
	2	86.3 ± 1.3	6.56×10−15
	3	78.9 ± 1.4	2.17×10−9
	4	82.2 ± 1.5	6.55×10−12
	5	81.0 ± 1.3	1.35×10−10
	Avg.	82.2 ± 2.7	4.30×10−10

Offline performances represented as classification accuracies estimated using 10 runs of stratified 10-fold CV. The classification accuracy is defined as the probability of correctly classifying a trial given the feature, f^⋆, i.e. $P\left(\text{correct}\right|f^{\star })=P(\mathcal{I}|f^{\star }\in \mathcal{I})P\left(\mathcal{I}\right)+P\left(\mathcal{W}\right|f^{\star }\in \mathcal{W}\left)P\right(\mathcal{W})$, where $P\left(\mathcal{I}\right|f^{\star }\in \mathcal{I})$ and $P\left(\mathcal{W}\right|f^{\star }\in \mathcal{W})$ are defined in Online calibration in the Methods section, and $P\left(\mathcal{I}\right)$ and $P\left(\mathcal{W}\right)$ are the prior probabilities of idling and walking class, respectively.