With the increasing availability of unobtrusive, and inexpensive sensors in smart environments, online sensor data segmentation becomes an important topic in reconstructing and understanding sensor data. Usually, in the literature, the segmentation is either performed by following a fixed or a dynamic time-window length. As stated in several works, static time-window length has several drawbacks while adjusting dynamically the window length is more appropriate. However, each of previous methods for dynamic data segmentation targets only a particular type of application. Hence, there is a need for a general method independent of applications providing high degree of usability. To achieve this aim, in this paper, we propose a novel method that dynamically adapts the time-window size. The proposal is designed to be applied in a wide range of applications (activity recognition, decision making, etc.) by combining statistical learning and semantic interpretation. This hybridization allows to analyze the incoming sensor data and choose the better time-window size. The presented approach has been implemented and evaluated in several experiments using the real dataset Aruba from the CASAS project