There is an increasingly urgent demand for new high-performance glass in the field of glass science and engineering. Traditional trial-and-error methods and physical modeling suffer from issues such as low efficiency, high cost, and insufficient accuracy. The emergence of artificial intelligence and machine learning has brought new breakthrough methods for glass design and development. Through dataset construction, model training, and validation, it can efficiently predict glass composition, structure, and performance. This paper elaborates on the basic principles of machine learning and core algorithms (including supervised and unsupervised learning), summarizes the application achievements of machine learning in various types of glass in recent years, and focuses on reviewing the research progress of composition-performance, composition-structure, and composition-structure-performance modeling and design of glass materials based on learning. Existing studies have shown that machine learning can significantly improve the accuracy of glass performance prediction and development efficiency, but it still faces challenges such as insufficient generalization ability and difficulty in fitting complex structures. In the future, with the improvement of technology and integration across multiple fields, machine learning will continue to promote innovative development in glass science and provide more efficient technical support for the research and development of new glass.