Ambient sensors of a smart home offer new services to occupants related to security, comfort and energy management. This makes it possible to estimate the practices of the occupants and determine energy consumption, by estimating the number of occupants per zone, their activities and routines, with a requirement for digital sobriety (i.e fewer sensors possible, but enough to ensure a service). The selection of a pertinent sensor configuration is an important aspect in human activity recognition (HAR). This problem includes the selection of the number and type of the sensors, as well as the identification of the most informative ones. In this paper, we propose a novel approach to determine the most informative sensor types (among motion detection sensors and door contact sensors) based on the criterion of mutual information. The selected configuration of sensors is applied to recognize daily activities in real-time in a real-world dataset: Aruba, from the CASAS project. The simulation results show good performance of classification and a save of processing time. Highlights • Consider the mutual information criterion to select the minimum setup of sensors • Recognize human activities in real-time with the selected setup of sensors. • Save of the processing time with a minimum setup of sensors.