One of the major problems faced by automated human activity recognition systems is the scalability. Since the probabilistic models employed in activity recognition require labeled data sets for adapting themselves to different users and environments, redeploying these systems in different settings becomes a bottleneck. In order to handle this problem in a cost effective and user friendly way, uncertainty sampling based active learning method is proposed. With active learning, it is possible to reduce the annotation effort by selecting only the most informative data points for annotation. In this paper, three different measures of uncertainty have been used for selecting the most informative data points and their performance have been evaluated by using real world data sets. It has been shown that the annotation effort can be reduced by a factor of two to four, depending on the house and resident settings in an active learning setup.