Open AccessA prediction model for superconductor critical temperature using stepwise discriminant analysis based on feature extraction
Author(s) -
Peng Guo,
Wei Li,
Z. Y. Su
Publication year - 2019
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1298/1/012020
Subject(s) - discriminant , principal component analysis , linear discriminant analysis , superconductivity , feature extraction , python (programming language) , cluster analysis , dimensionality reduction , computer science , pattern recognition (psychology) , artificial intelligence , materials science , data mining , algorithm , mathematics , physics , condensed matter physics , operating system
Although the critical temperature is a very important step for the extensive application of superconductors, it’s difficult to find the correlations of many kinds of physical superconductor properties and guarantee the accuracy in predicting the critical temperature. In this paper, an efficient prediction model using stepwise discriminant analysis based on feature extraction is provided to give the reduction of superconductor physical properties and predict the superconductor critical temperature. Firstly principal component analysis and clustering analysis are implemented to reduce the data dimension of superconductor physical properties and give the reductive clusters to complete the feature extraction. The 71 physical properties data of 1300 superconductors is efficiently reduced to 3 main components and 7 clusters. According to the extracted features and improved stepwise discriminant analysis based on the principle of binary search, bayesian discriminant function of each layer is established. At last, Python programming is designed to input the characteristic values and output the predicted the range of critical temperature to finish the efficient computer implementation of this model. Its result that only to the fifth layer, the superconductor critical temperature accuracy of 20,000 × 71 data matrix is 3. 125 has verified the efficiency of this model.