Mobile Information Systems / 2016 / Article / Tab 11

Research Article

A Mobile Application for Easy Design and Testing of Algorithms to Monitor Physical Activity in the Workplace

Table 11

Classifier performance when trained with generalized walking data.
(a) Performance by cumulative indexes

Correctly classified instances 92.68%
Incorrectly classified instances 7.32%
Kappa statistic 0.829
Mean absolute error 0.034
Root mean squared error 0.143
Relative absolute error 23.16%
Root relative squared error 52.71%

(b) Confusion matrix

Classified as
WalkingJoggingUpstairsDownstairsSittingStanding

Walking 10221 15 125 116 15 17
Jogging 46 1662 7 9 0 0
Upstairs 253 33 385 70 0 2
Downstairs 231 6 53 328 0 0
Sitting 15 0 1 1 366 7
Standing 14 0 0 0 12 316

(c) Detailed accuracy by class

Precision Recall-scoreClass

0.948 0.973 0.96 Walking
0.969 0.964 0.966 Jogging
0.674 0.518 0.586 Upstairs
0.626 0.531 0.574 Downstairs
0.931 0.938 0.935 Sitting
0.924 0.924 0.924 Standing

0.921 0.927 0.923 Weighted average