An Efficient Normalized Rank Based SVM for Room Level Indoor WiFi Localization with Diverse Devices
Table 3
Percentage of accurately estimated locations based on linear SVM adopting multiple combinations of constraints violation along with comparison between RBF kernel and regularized Logistic Regression; the radio map is defined upon the proposed normalized rank transformation.
Device type
Classification type
RBF kernel
L2-R L2-N
L2-R L2-N
L2-R L1-N
L2 LR
L1-R
L1
(primal)
(dual)
(dual)
(primal)
L2-N
LR
Oppo A31c
BBK
98.75%
98.75%
98.75%
99.17%
95.00%
Coolpad 8730L
100%
100%
100%
100%
96.87%
99.58%
99.79%
Gionee
100%
100%
100%
100%
96.04%
94.79%
HTC One E8
98.75%
98.75%
98.75%
98.75%
91.04%
95.00%
Huawei GRA-CL00
100%
100%
100%
100%
98.12%
97.08%
Lenovo A788t
99.79%
99.79%
99.79%
100%
96.87%
98.54%
Meizu
100%
100%
100%
100%
94.58%
100%
100%
Oppo R7c
99.58%
99.58%
99.58%
100%
95.62%
Xiaomi
99.17%
99.17%
99.17%
99.17%
Xiaomi Cancro
100%
100%
100%
100%
91.25%
Samsung klteduoszn
100%
100%
100%
100%
97.29%
98.75%
Meizu M2 note
100%
100%
100%
100%
91.25%
94.37%
100%
Samsung trlteduosctc
99.58%
99.58%
99.58%
99.58%
92.08%
95.83%
98.95%
BBK Vivo
100%
100%
100%
100%
93.95%
98.33%
Standard device accuracy
100%
100%
100%
100%
94.79%
99.37%
97.50%
Recall (ratio)
1
1
1
1
0.94
0.99
0.97
Precision (ratio)
1
1
1
1
0.91
0.99
0.98
() R stands for regularization; () N stands for norm; ( ) LR stands for logistic regression. We put in bold the percentages lower than 90% for easiness of observation.
