University of Twente Student Theses
A Markovian approach to mobile user classification with semantic location data in mobility chains
Zwienenberg, Jop (2022) A Markovian approach to mobile user classification with semantic location data in mobility chains.
|
PDF
1MB |
| Abstract: | The large amount of mobility behavior data of people collected by companies can provide insights into how people move over the day. The software of Mobidot captures via GPS 24/7 the mobility behavior of Dutch individuals and groups over the whole world via their mobile phones. Mobidot has a rule-based method to determine the purposes of trips during the activity recognition phase. Examples of activities are working, shopping, and leisure. The company’s ultimate overall goal is to find a method to build a complete user profile which serves as an input to an improved activity recognition via the rule-based method. The improved activity recognition would provide valuable answers to Mobidot’s customers for market research, panel surveys, urban policy development, and impact evaluation. As an example, the data can be used as an input for government policies to achieve a smoother and better distributed traffic flow. In this thesis, we work partially towards this goal by developing two different binary classification models for classifying two types of users, users with and without a job. We use the discrete time Markov model (DTMC) and the long short-term memory (LSTM) neural network for this. The choice for this specific binary classifier is - sort of - arbitrary, yet serves as a building block for more sophisticated and more extensive classifiers. For training the classifiers, we use a user’s daily semantic location sequences over a certain period with known job status as input. It is shown that a DTMC performs well in classifying these types of users by comparing its performance in the form of the balanced accuracy with the performance of the baseline LSTM neural network. Classification is done by calculating the Jensen-Shannon distance between users, which can be interpreted as a distance between their Markov models. The outcomes of this thesis suggest that DTMC models can be a useful building block in the design of methods to generate user profiles. |
| Item Type: | Essay (Master) |
| Faculty: | EEMCS: Electrical Engineering, Mathematics and Computer Science |
| Programme: | Applied Mathematics MSc (60348) |
| Link to this item: | https://purl.utwente.nl/essays/92939 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page