Where you go is who you are: a study on machine learning based semantic privacy attacks

Journal of Big Data

Table 1 Comparing attack-scenarios. For the place categorization task, the accuracy over location-visit events is reported. User-profiling error, user identification accuracy and privacy loss measure the success of the attacker in user-profiling. The table only shows the results for an obfuscation radius of \(r=100\)

Check-in data	POI data	Method	Place categorization	User profiling
Check-in data	POI data	Method	Accuracy	Profiling error	Top-5 identifi-cation accuracy	Privacy loss (median)
Foursquare (NYC and Tokyo)	Foursquare	Random	0.159	0.235	0.003	1.000
		XGB (temporal)	0.344	0.205	0.023	1.112
		spatial join	0.459	0.208	0.183	2.296
		XGB (spatial)	0.580	0.140	0.328	5.574
		XGB (spatio-temporal)	0.616	0.124	0.404	11.001
	OSM	Random	0.157	0.235	0.002	1.000
		Spatial join	0.224	0.341	0.013	1.051
		XGB (temporal)	0.328	0.206	0.022	1.116
		XGB (spatial)	0.476	0.172	0.117	1.949
		XGB (spatio-temporal)	0.503	0.157	0.160	2.316
Yumuv study (Switzerland)	Foursquare	Random	0.370	0.367	0.010	1.000
		Spatial join	0.453	0.698	0.009	1.000
		XGB (spatial)	0.541	0.331	0.034	1.095
		XGB (temporal)	0.553	0.333	0.038	1.108
		XGB (spatio-temporal)	0.618	0.293	0.065	1.225
	OSM	Random	0.372	0.367	0.010	1.000
		Spatial join	0.384	0.698	0.009	1.000
		XGB (spatial)	0.541	0.332	0.034	1.093
		XGB (temporal)	0.553	0.333	0.038	1.108
		XGB (spatio-temporal)	0.623	0.289	0.060	1.256