Reducing the energy consumption to the minimum is a crucial design requirement for all body area sensor networks. Sensors deployed on the human body, especially at the limbs often move along different positions. Usually, the transmit power is set to a sufficiently high value to achieve reliable transmission for the constellation with highest attenuation. For periodic movements, data transmission can be carried out at the position of the lowest path loss between the sender and the receiver, provided this position can be reliably identified. We propose a novel framework that predicts this position using acceleration data and the received signal strength. By learning a correlation between these signals, accurate predictions can be performed and up to 24.7% of the power spent by a Bluetooth Low Energy module for the transmission of a packet can be saved while still achieving the same packet error rate as with sending using the higher transmit power.