Dementia affected 47 million people worldwide in 2016, which caused $ 818 billions of estimated costs in total. Due to high costs and inadequate governmental support, over 90 % of people with dementia in low- and middle-income countries need to be cared for at home. However, the multitude of implications overburdens both the people affected as well as their caregivers. Existing health monitoring systems for ambulatory care only offer dementia-specific support functions in combination with a variety of sensors, applications and administration efforts. Therefore, we present a method to support ambulatory care with simple ambient sensor settings. Our approach is intended to model the behaviour of the resident to derive habits and possible anomalies regarding dementia. To support planning of care measures and detect dementia onset, we start with day-night rhythm and night-time activity as relevant parameters, since they are associated with dementia and are beneficial for the implementation of care. Our primary objective is to automate the personalization of existing monitoring systems with as few ambient sensors as possible. The challenge of this work is to learn these parameters from a brief sensor data history about people who are living alone, are unable to handle wearable devices and cannot give autonomous feedback on their activities of daily living.