Semantic Labeling of User Location Context Based on Phone Usage Features

Table 4

Comparison of data and solutions.

Solution	Users	Percentage of cases			Labels	Features	Best; classifier	Accuracy (%)
		Home	Work	Other				Overall	Home	Work
[6]	80	25	30	45	10	2,769,200	GBT	75.1	N/A	N/A
[7]	80	25	30	45	10	54	()	65.8	87	85
[8]	80	25	30	45	10	1177	()	73.3	100	100
[9]	114	25	29	46	10	500	()	75.5	92	90
#1 places	114	29	26	45	10	3 (SFS)	BT	68.5	92	90
#1 cum. s.	114	31	29	40	10	9	BT	68.4	94	92
#1 visits	114	52	38	10	3	14	NN	76.7	83	86
#1 places	114	36	33	31	3	14	LR	89.2	93	92
#1 cum. s.	114	36	35	29	3	9	LR	89.5	92	88
#2 cum. s.	16	31	26	42	3	11	LR	85.9	81	83

Multilevel 2-method (SMO and simple logistic), fusion with decision tree. Ensemble of binary classifiers using 1NN and SVM. Combination of multiclass random forests and one-versus-all random forest binary classifiers.