This work presents an approach to automatically adapt domestic hot water heaters to to individual human behavior based on real IoT data. For this purpose, a large collection of data from domestic hot water heaters is analyzed to learn the consumption behaviors of each user. The human behavior is learned using two different approaches that we compare: neural networks and Gaussian processes with periodic kernels. The learned human behavior is used to create an optimal hot water schedule that adapts to each user and thus saves between 20 and 34% of the energy used with a default schedule. We also propose an eco-parameter so that each user can determine a trade-off between maximum comfort (always having hot water available) and maximum energy savings.