Artificial neural network has becoming a mainstream technology in the domain of complex materials data analysis. Based on a slag glass-ceramic system we brought forward a virtual sample technology to increase the training samples by fluctuating the content of main compositions in a proper small amplitude. Simulation results proved that a good virtual sample set can not only improve the network's prediction ability considerably, but can also suppress the "overtraining" phenomenon. Therefore a virtual sample improved neural network model can learn the relationship from a small size experimental data set and give an accurate and stable prediction for the test samples. This is more helpful to the material data analysis and can facilitate the design and development for new materials.

Keywords: Artificial neural network; Material Data Analysis; Virtual sample technology; Slag Glass-Ceramics