Join the CRAWDAD community

Reset your CRAWDAD account password

Datasets and tools by name

Datasets and tools by release date

Datasets and tools by keyword:

Datasets by measurement purpose:

About the CRAWDAD project

CRAWDAD references in CiteULike

CRAWDAD contributors by country

CRAWDAD members by country


CRAWDAD sponsors

Related websites


 Search CRAWDAD via Google:

The yonsei/lifemap dataset (v. 2012-01-03)  >  the mobility traceset

There is 1 trace in this traceset
download the file
from a CRAWDAD mirror:  US UK AU
size="22MB" md5="b9c6afa32209b616f13bef5d815711ca" type="zip"
last modified
reason for most recent change
the initial version.
short description

Mobility data collected by LifeMap monitoring system at Yonsei University in Seoul.


We deployed our mobility monitoring system, called LifeMap, to collect fine-grained mobility data from commercial mobile phones over two months in Seoul, Korea.

The dataset contains location information (latitude and longitude) with accuracy (error bound), Wi-Fi fingerprints (MAC address and signal strength of surrounding Wi-Fi APs), user-defined types of places (workplace, cafeteria, etc.). Our system continuously collected this information every 2 to 5 minutes for everyday location monitoring.

release date
date/time of measurement start
date/time of measurement end

The first level of sensing uses GSM to obtain the Location Area Code (LAC) to detect exceptions within the predicted sensing schedule. The first level continuously monitors the LAC with minor energy consumption, since a mobile phone basically updates the LAC for voice communication. The system does not activate the second level until the next sensing time, if the observed LAC follows a predicted sequence-pattern. Otherwise, if the first level detects an exception, the system immediately uses the second level to collect a new pattern of individual mobility. For example, if a user normally goes to the office on weekday mornings, SmartDC only turns on Wi-Fi near the entrance of the office if the expected LAC is observed. If a user goes on a business trip, the system uses the mobility learner when it detects a new pattern of LAC. The second level uses Wi-Fi scanning to recognize a change of places and revisited places. The basic operation is that if a user is stationary, the signal fingerprints of surrounding Wi-Fi APs are relatively similar to each other. We use a scan window to perform multiple scans to tolerate noisy signals. The system generates meaningful places when it detects the stationary state. When a user revisits a place, the system aggregates mobility data and reuses physical location information without activating additional sensors. The second level also uses wireless communication to obtain coarse location from the WPS provided by Android. The third level activates GPS to acquire fine location, if the system fails to get accurate location in the second level.

The time interval used for data collection is set to two minutes. The Wi-Fi scanning intervals and window size are 10 seconds and 30 seconds respectively, and the similarity threshold of the Wi-Fi vector is set to 0.7. The accuracy threshold for the WPS is set to 500 meters. The GPS is activated for 30 seconds for single positioning, which is common in GPS usage. To reduce computation overheads in adaptive duty cycling, we (1) converted the float value to an integer value with 10-3 precision, (2) used discrete time intervals in minutes, and (3) scaled down the energy budget, dividing it by the energy consumption of Wi-Fi scanning, which is the minimum cost in our scheme. We set the energy budget E and the sensing cost c as follows: The maximum energy budget is (1,400 mA×3.7 V34.5 mW)×3,600 s=18.5 kJ, which is the available battery capacity excluding the energy cost of idle state. If the battery level and energy constraints are 30% and 10%, the allowed energy budget is 18.5 kJ×0.3×0.1=555 J. The level 2 sensing cost is the energy consumption of Wi-Fi scanning: 114.5 mW×30 s=3.5 J. The level 3 sensing consumes 3.5 J+440.8 mW×30 s=16.7 J: the energy of level-2 sensing and reading GPS for 30 seconds.


We removed personal information such as user names and users' MAC addresses. User-labelled place names are anonymized with one letter of alphabet (A, B, C, and so on) and number (from 001 to 999). We anonymized MAC addresses of APs by changing the second and the fifth parts of the MAC address into randomized characters (e.g., 00:00:00:00:00:00 is changed into 00:xx:00:00:xx:00). The uniqueness of APs is maintained: the same address indicates the same AP.


The users spent some of their time in regions without Wi-Fi coverage.

 the yonsei/lifemap/mobility/2011 trace
 how to cite this traceset