Semantic Labeling of User Location Context Based on Phone Usage Features

Figure 11

Rejecting samples with low classification confidence. T: confidence threshold, R: ratio of rejected samples to all samples, I: accuracy improvement obtained by rejecting low confidence samples, and A: classification accuracy within the samples with confidence threshold. Data representations: data set #1, (a) visits, (b) places, and (c) cumulative samples; (d) data set #2, cumulative samples. Accuracies given as ratios (unitless).

(a)

(b)

(c)

(d)