The use of artificial neural networks in materials science based research

The use of computer modelling techniques is extensive in scientific research. Artificial neural networks are now well established, and prominent in the literature, when computational based approaches are involved. The materials science and engineering research community has and continues to take advantage from new developments in these areas with different applications regularly emerging, along with the degree of sophistication utilised. However, with this increased use there is unfortunately a growing tendency for the misapplication of neural network methodologies, limiting their potential benefit. Central to the problem is the use of over complicated networks that are frequently mathematically indeterminate, and by using limited data for training and testing. This problem is not unique to one particular field, but has prompted the authors to bring it to the attention of the materials research community in order to elaborate the issues and hopefully prevent others and potentially new researchers from continued misuse of neural networks in the future. (C) 2007 Elsevier Ltd. All rights reserved.