This paper presents a behavior description algorithm from time-series data on daily life activities extracted using home sensors. In a previous work, we proposed a method of time-series data clustering based on hidden Markov models (HMMs). This method separates the time-series data in segments of equally short length, and applies a behavior label for each segment. However, the change points of behaviors are not clear and it is difficult to detect short length behavior. In this paper, we propose a new behavior description algorithm by introducing singular spectrum transformation (SST), a nonlinear transformation used for change-point detection, and apply it to our previous method. This method enables more precise change-point detection and behavior labeling.