Learning the Structure of Graphical Models Based on Discrete Time Series Data in the Context of Ambient Assisted Living

This work addresses the issue of building a probabilistic system in an ambient assisted living environment to ensure a proper living for older adults. The focus lies on the early prediction of human activities based on domotic sensor data which form a temporalsequential data set. In contrast to commonly used methods in sequential data mining, data in hidden streams and with variable temporal spans are considered. The aim in this context is to detect recurrent patterns in a stream of domestic sensor data using the TemporalPattern (T-Pattern) algorithm and to automatically generate probabilistic finite-state automata.