This study includes the use of machine learning techniques for occupancy detection in smart buildings. One of the most important elements in smart buildings is the effective use of energy. For this reason, a machine learning technique which required the minimum amount of hardware was investigated in order to realise effective energy use while performing occupancy detection. For this purpose, the machine learning technique that provided the highest accuracy using the minimum number of features in a sample dataset was tested. To reduce the classification performance and energy requirement of the information collected from different environmental sensors in the study, time requirements were analysed and presented. Among the implemented machine learning algorithms (support vector machine (SVM), k-nearest neighbour (KNN), decision trees) the highest performance (with minimum number of features) was achieved by using the indoor air temperature and relative humidity features in the dataset and 99.5123% accuracy was obtained.