Doctoral defence: Toivo Vajakas "Towards integration of mobile network data into analyzing human mobility“

On 28 November at 12:15 Toivo Vajakas will defend his thesis "Towards integration of mobile network data into analyzing human mobility“ to obtain the degree of Doctor of Philosophy (in Computer Science). 

Supervisors
Professor Eero Vainikko, University of Tartu
Associate Professor Amnir Hadachi, University of Tartu

Opponents
Professor Sidharta Gautama, Ghent University (Belgium)
Associate Professor Claudio Roncoli, Aalto University (Finland)

Summary
The mobile network operators continuously produce data that can be used to derive the approximate location of the people. This is called passive mobile positioning. 

The passive mobile positioning data have very good population coverage. These data are used in spatial planning, location-based advertising, and mobile operator benefits from localization of reported network malfunction. One big problem is the large uncertainty of the derived location. 

Passive mobile positioning data contains a significant percentage of strongly erroneous location estimates. The real location is strongly correlated to other measurements - location changes continuously, and people tend to repeat the same patterns on different days, and different phones behave similarly in the same circumstances. The work is based on the hypothesis that the correlations can be used to distinguish the "reasonable" measurements from outliers. 

In work proposes a method to use the statistics over all mobile phones to assess the probability of movement for a given mobile phone. This is useful when the phone is connecting to another antenna, to reduce the so-called ping-pong effect -- even a stationary mobile phone can seem to move when switching the antennas. 

The probability of a phone connecting to the given antenna is influenced by other available antennas nearby. A probabilistic Bayesian model was developed to account for this effect. 

In mobility analysis, it is important to distinguish stop and move episodes of observed devices. The author proposed improvements to the prior SKF algorithm. 

The author developed a method to test the hypothesis that part of positioning problems stem from incorrect cell azimuth data azimuths. The hypothesis was tested and the test data did not exhibit incorrect cell data. 

Methods were found to improve the result in certain situations, but no universal "silver bullet" was found to universally improve the passive mobile positioning. 

The defence will be held also in Zoom (meeting ID: 975 8871 8901, passcode: ati).