Abstract: In recent years, MAX phase crystals have become one of the global research hotspots, because of their unique nanolayered crystal structure with its advantages of self-lubrication, high tenacity and electrical conductivity.M2AX phase crystals have the properties of ceramic and metal compounds, and have thermal shock resistance, high tenacity, electrical conductivity and thermal conductivity.Due to the difficulty of preparing single-phase samples for such materials,the development of this materials is limited.Active learning is a machine learning method that uses a small number of labeled samples to achieve high prediction performance. In this paper, we combine the efficient global optimization and the residual active learning regression to propose a improved active learning selection strategies, called RS-EGO, and proposed for predicting the bulk modulus, Young's modulus, shear modulus and Poisson's ratio on a dataset of 169 M2AX phase crystals. It is found that the improved selection strategy RS-EGO has better prediction performance for the four objectives and also more suitable for the material performance prediction problem for small data samples.