In the near future, smart home systems will play more and more important role to provide comfortable and safe life to human. Today, we already have some realistic way to monitor the daily life of human and recognize their activities by cameras or wireless sensing technology. However, the current research still faces the challenge to the prediction of human activities. In this paper, we analyse the similarity between human activities of daily living and deep neural networks. Inspired by this, the paper proposes a method to predict human activity by deep learning model and evaluates the performance of the approach with real world data. Compared with the traditional algorithm, our approach reaches higher prediction accuracy. In the future, we will try to improve the prediction accuracy and add more kinds of activities.