The population age requires assistant systems to assist the elderly to live in a familiar place as long as possible. In the wide range of the smart home applications, abnormal behaviour detection is attracting researchers due to its important benefits for the safety of the elderly people. In this research, a hybrid approach to description logic learning is proposed to learn normal behaviours of the elderly in smart homes. Negation As Failure (NAF) can be later used to detect abnormalities based on the learned rules. In addition, a methodology for generating context-awareness smart home datasets based on use cases is also proposed to evaluate the learning algorithm. The experimental results show that the proposed algorithm is suited to this problem. The learning speed and scalability of the proposed algorithm are significantly better than other description logic learning algorithms used in the comparison.