User identification plays a vital role in smart home applications such as home automation and intruder detection. Prior work has shown that multiple occupation, privacy and obtrusiveness has become limitation. Passive identification using privacy-preserving sensors aims at enhancing the user experience.

In this thesis, we propose a user identification system using capacitive touch sensors to monitor residents’ touch interactions with objects in a home environment. This approach is based on the physical phenomenon that different residents interact with objects in different ways. The system extracts richer information content from the sensors and turns it into a supervised learning classification problem.

To evaluate the system we conduct an 11-day experiment in a real home kitchen by deploying touch sensors on 19 distinct objects. Furthermore, we experiment with how different smoothing methods affect the accuracy of touch detection and presence detection in the system. As classification algorithms Support Vector Machine, Classification and Regression Tree and Random Forest were studied. We also explore how different window lengths affect the accuracy of the system.