Bayesian Influence Diagram models are an extension of Bayesian probabilistic graphical models that includes Action nodes and utility costs.  We designed and trained a Bayesian graphical model to recommend actions to elders (and their families) to improve sleep.  One set of input features came from in-home atmospheric sensors, including temperature, humidity, pressure and CO2.  Another set of input features came from a bed/sleep sensor placed under the mattress, including time in bed awake and in REM, light, medium and deep sleep.