The methods of machine learning based on the data science can deal with the corresponding studies in different disciplines based on the data accumulated in theory and experiments. Machine learning promotes the development of data-intensive scientific discoveries, thus making it a "fourth paradigm" that leads to the related scientific research after "theory, calculation, and experimentation". Among different materials, perovskite material has some unique advantages of rich composition, adjustable band gap, and broad development space, but this material does not reach the practical standards such as environmental friendliness in applications. Therefore, the exploration of perovskite material and its applications based on machine learning can accelerate the discovery of novel perovskite material, and explore the relationship between the physical and chemical characteristics of perovskite material, therefore providing a guidance for the development of environmentally friendly high-performance perovskite devices. This review represented the research process of machine learning for perovskite material, summarized some research work on machine learning in perovskite material properties and device exploration, and discussed the existing difficulties and challenges. In addition, the future development direction and trend were also prospected.