On April 26th, 2022, Valentino Servizi successfully defended his Ph.D. thesis entitled “Mining user transport behavior from smartphones.”
Examiners were Prof. Filipe Rodriguez (DTU), Trude Tørset (NTNU), and Fang Zhao (MMM).
The defense session was chaired by Prof. Thomas Kjær Rasmussen (DTU).
Find below the abstract of the thesis! For more information, feel free to contact Valentino at firstname.lastname@example.org or directly via his Linkedin page.
This Ph.D. thesis contributes to enabling high-resolution measures of human transport behavior variations from smartphones. Smartphones can contribute to yielding the most prosperous perspective on the study of transport behavior variations both between and within users. While traditional approaches are already measuring behavior variations between users, we need higher resolution to measure these variations within the same user. However, handling such a higher resolution provides a new complex set of challenges.
We pinpoint and examine the problems limiting prior research up-front, exposing drivers to intuitively rank relevant machine-learning algorithms, identify physical limitations, and cast a relationship among human/system interactions, methods, and data. Next, we focus on two fundamental binary classification problems centered on Geographic Positioning System (GPS) trajectories. Both underpin many current and upcoming smartphone-based technologies deployed to measure human transport behavior variations: one problem is “stop” classification; the other is presence detection inside the transport network.
The solution combines GPS time series fused with spatial context information for the first problem. For the second problem, the solution exploits GPS and Bluetooth Low Energy technology. Both solutions rely on the extension of several artificial neural network frameworks based on the back-propagation algorithm. We also study the sensitivity of these methodologies to noise in both sensor signal and ground truth quality. This work underpins novel solutions reducing the dependency on labels and improving comparability across methods.