Location Estimation with Wi-Fi Fingerprint Data
Tárkányi, Hanga Zsófia (2025)
2025
Master's Programme in Computing Sciences and Electrical Engineering
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
Hyväksymispäivämäärä
2025-06-30
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202506277455
https://urn.fi/URN:NBN:fi:tuni-202506277455
Tiivistelmä
Wi-Fi fingerprint data-based indoor localization has been a popular topic in both academic and industrial research, but the difficulties associated with its real-life applications leave many open questions to this day – especially for large and complex locations – for instance with regards to effective data collection, computation, and handling changing environments.
Unlike other works – that most often attempt to predict specific coordinates – this thesis investigates the feasibility of a classification of rooms or zones of a building, and aims to find the method that is best suited for this task. For this project a specific dataset collected throughout two floors of a university building in Spain was used.
To investigate the classification problem, multiple experiments with machine learning methods – such as support vector machines – were performed, and the accuracy of the models and the stability of their predictions were observed after different model configurations had been applied – such as different types of encoding, and different kernels – as well as through experiments with noise simulation.
As the maintenance of information security is essential for the data transfer in the use case that motivated this project, the effects of a secure, hashed data encoding were also tested.
The results have shown that high accuracy is achievable for the classification (over 90% overall accuracy for some methods), even with a noisy Wi-Fi fingerprint dataset. The secure hashed encoding does not noticeably decrease the accuracy (less than 0.5% drop). The best classification method for this data is the linear kernel support vector machine, whereas weighting the data by signal strength did not show improvement.
Unlike other works – that most often attempt to predict specific coordinates – this thesis investigates the feasibility of a classification of rooms or zones of a building, and aims to find the method that is best suited for this task. For this project a specific dataset collected throughout two floors of a university building in Spain was used.
To investigate the classification problem, multiple experiments with machine learning methods – such as support vector machines – were performed, and the accuracy of the models and the stability of their predictions were observed after different model configurations had been applied – such as different types of encoding, and different kernels – as well as through experiments with noise simulation.
As the maintenance of information security is essential for the data transfer in the use case that motivated this project, the effects of a secure, hashed data encoding were also tested.
The results have shown that high accuracy is achievable for the classification (over 90% overall accuracy for some methods), even with a noisy Wi-Fi fingerprint dataset. The secure hashed encoding does not noticeably decrease the accuracy (less than 0.5% drop). The best classification method for this data is the linear kernel support vector machine, whereas weighting the data by signal strength did not show improvement.