Methods and systems for occupancy prediction using historical occupancy patterns are described. In an embodiment, an occupancy probability is computed by comparing a recent occupancy pattern to historic occupancy patterns. Sensor data for a room, or other space, is used to generate a table of past occupancy which comprises these historic occupancy patterns. The comparison which is performed identifies a number of similar historic occupancy patterns and data from these similar historic occupancy patterns is combined to generate an occupancy probability for a time in the future. In an example, time may be divided into discrete slots and binary values may be used to indicate occupancy or non-occupancy in each slot. An occupancy probability for a defined future time slot then comprises a combination of the binary values for corresponding time slots from each of the identified similar occupancy patterns.