This paper explores the methods that can estimate a surrounding symptom level of dementia patients by monitoring the sleep features such as the sleep rhythm (cycle) and the light/deep sleep ratio. For this purpose, we employ Watanabe's method to estimate the sleep stages as the sleep feature, which can be transformed from the heartbeat data acquired by the mattress-based sensor. To explore the dementia estimation methods, we conducted the human subject experiment in the hospital. The analysis of the sleep stage of the dementia patient who has the light cognitive function impairment with repeating good and bad healthy conditions have revealed that the averaged cycles of the REM sleep and the light/deep sleep ratio have a great potential of the evaluation criteria that can estimate a surrounding symptom level of dementia patients.