Linear SVM Models for Online Activity Recognition - Smartphone-Based Human Activity Recognition - page 103

Table 6.6 Confusion matrices obtained with D 2 T and the MC-L1-L2-SVM algorithm using dif-
ferent values of as ʻ

Activity WK WU WD SI

ST LD Activity WK WU WD SI ST LD

MC-L1-SVM MC-L1-L2-SVM ʻ = 0 . 001

WK 492 3 1 0 0 0 WK 493 2 1 0 0 0

WU 18 452 0 1 0 0 WU 16 455 0 0 0 0

WD 5 1 413 1 0 0 WD 5 2 413 0 0 0

SI 0 1 1 436 53 0 SI 0 1 0 436 54 0

ST 0 0 0 15 517 0 ST 0 0 0 14 518 0

LD 0 0 0 0 0 537 LD 0 0 0 0 0 537

Accuracy 96.61% Accuracy 96.78%

MC-L1-L2-SVM ʻ = 0 . 005 MC-L1-L2-SVM ʻ = 0 . 01

WK 493 2 1 0 0 0 WK 493 2 1 0 0 0

WU 15 456 0 0 0 0 WU 15 456 0 0 0 0

WD 5 2 413 0 0 0 WD 4 2 413 1 0 0

SI 0 2 0 434 55 0 SI 0 2 0 436 53 0

ST 0 0 0 14 518 0 ST 0 0 0 14 518 0

LD 0 0 0 0 0 537 LD 0 0 0 0 0 537

Accuracy 96.74% Accuracy 96.81%

MC-L1-L2-SVM ʻ = 0 . 05 MC-L1-L2-SVM ʻ = 0 . 1

WK 493 2 1 0 0 0 WK 493 2 1 0 0 0

WU 15 456 0 0 0 0 WU 17 454 0 0 0 0

WD 4 2 413 1 0 0 WD 5 2 412 1 0 0

SI 0 1 0 437 53 0 SI 0 1 0 436 54 0

ST 0 0 0 12 520 0 ST 0 0 0 13 519 0

LD 0 0 0 0 0 537 LD 0 0 0 0 0 537

Accuracy 96.91% Accuracy 96.74%

MC-L1-L2-SVM ʻ = 0 . 5 MC-L1-L2-SVM ʻ = 1

WK 493 2 1 0 0 0 WK 494 2 0 0 0 0

WU 18 453 0 0 0 0 WU 20 451 0 0 0 0

WD 3 1 414 1 1 0 WD 8 0 412 0 0 0

SI 0 3 0 433 55 0 SI 0 3 0 428 60 0

ST 0 0 0 12 520 0 ST 1 0 0 15 516 0

LD 0 0 0 0 0 537 LD 0 0 0 0 0 537

Accuracy 96.71% Accuracy 96.30%

MC-L2-SVM
WK

494 2 0 0 0 0

WU 20 451 0 0 0 0

WD 8 0 412 0 0 0

SI 0 3 0 428 60 0

ST 1 0 0 15 516 0

LD 0 0 0 0 0 537

Accuracy 96.30%