Skip to main content
. 2020 Feb 25;14:155. doi: 10.3389/fnins.2020.00155

Table 3.

Comparison of SVM accuracy achieved by the proposed bag-level representation.

Subject TSGSP SFBCSP SFTOFSRC Proposal Proposal*
A08T 95.8 97.0 ± 2.9 96.9 ± 3.4 99.0 ± 1.7 99.2 ± 1.6
A09T 81.3 97.8 ± 3.1 94.6 ± 3.4 98.7 ± 2.1 97.3 ± 3.2
A03T 93.8 98.8 ± 1.7 98.5 ± 1.9 97.7 ± 2.6 99.2 ± 1.6
A01T 87.0 91.8 ± 4.7 91.8 ± 3.9 96.0 ± 1.1 94.8 ± 3.5
A05T 90.4 90.6 ± 3.7 95.7± 2.1 93.0 ± 3.1 95.3 ± 4.4
A07T 91.4 94.7 ± 6.1 76.3 ± 5.7 88.0 ± 4.5 96.2 ± 4.6
A06T 63.9 67.9 ± 6.9 71.0 ± 6.4 74.5 ± 4.8 72.4 ± 7.9
A04T 74.3 63.5 ± 10.6 69.0 ± 7.1 70.3 ± 6.8 69.5 ± 9.3
A02T 64.7 58.4 ± 8.3 62.8 ± 5.9 64.0 ± 5.9 66.2 ± 5.3
Average 82.5 84.5 ± 5.3 84.0 ± 4.4 86.8 ± 3.5 87.8 ± 4.6

Each method performing the best individual accuracy is marked in bold. Abbreviation Proposal denotes the enhanced representation without optimizing the t-f atoms, while notation * includes this procedure. Each underlined subject achieves confident differences of performance with either proposal version.