A common assumption in activity recognition is that the system remains unchanged between its design and its posterior operation. However, many factors affect the data distribution between two different experimental sessions. One of these factors is the potential change in the sensor location (e.g. due to replacement or slippage) affecting the classification performance. Assuming that changes in the sensor placement mainly result in shifts in the feature distributions, we propose an unsupervised adaptive classifier that calibrates itself using an online version of expectation-maximisation. Tests using three activity recognition scenarios show that the proposed adaptive algorithm is robust against shift in the feature space due to sensor displacement and rotation. Moreover, since the method estimates the change in the feature distribution it can also be used to roughly evaluate the reliability of the system during online operation.